From bd689f91ac13c9f5fcfb74b9409e972f011b3a65 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Thu, 13 Jun 2019 14:43:23 +0300
Subject: [PATCH] Added TridentNet (state-of-art on MS COCO)

---
 build/darknet/x64/cfg/resnet152_trident.cfg | 2177 +++++++++++++++++++
 build/darknet/x64/partial.cmd               |    2 +
 cfg/resnet152_trident.cfg                   | 2177 +++++++++++++++++++
 include/darknet.h                           |    1 +
 src/coco.c                                  |    4 +
 src/conv_lstm_layer.c                       |   22 +-
 src/convolutional_layer.c                   |  119 +-
 src/convolutional_layer.h                   |    2 +-
 src/crnn_layer.c                            |    6 +-
 src/layer.c                                 |   52 +-
 src/parser.c                                |   15 +-
 11 files changed, 4493 insertions(+), 84 deletions(-)
 create mode 100644 build/darknet/x64/cfg/resnet152_trident.cfg
 create mode 100644 cfg/resnet152_trident.cfg

diff --git a/build/darknet/x64/cfg/resnet152_trident.cfg b/build/darknet/x64/cfg/resnet152_trident.cfg
new file mode 100644
index 00000000..c7e0e255
--- /dev/null
+++ b/build/darknet/x64/cfg/resnet152_trident.cfg
@@ -0,0 +1,2177 @@
+[net]
+# Testing
+#batch=1
+#subdivisions=1
+# Training
+batch=64
+subdivisions=64
+width=608
+height=608
+channels=3
+momentum=0.9
+decay=0.0005
+angle=0
+saturation = 1.5
+exposure = 1.5
+hue=.1
+
+
+learning_rate=0.001
+burn_in=1000
+max_batches = 10000
+
+policy=sgdr
+sgdr_cycle=1000
+sgdr_mult=2
+steps=4000,6000,8000,9000
+#scales=1, 1, 0.1, 0.1
+
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=7
+stride=2
+pad=1
+activation=leaky
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+
+# Conv 4
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+
+
+
+### TridentNet - large objects - Start
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+
+## Conv 5
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=512
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=1
+stride=1
+pad=1
+filters=2048
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=24
+activation=linear
+
+[yolo]
+mask = 8,9,10,11
+anchors = 8,8, 10,13, 16,30, 33,23,  32,32, 30,61, 62,45, 59,119,   80,80, 116,90, 156,198, 373,326
+classes=1
+num=12
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=0
+
+### TridentNet - large objects - End
+
+
+
+
+
+
+
+### TridentNet - medium objects - Start
+
+[route]
+layers = 165
+
+[convolutional]
+share_index=166
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=167
+dilation=2
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=168
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=170
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=171
+dilation=2
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=172
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=174
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=175
+dilation=2
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=176
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=178
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=179
+dilation=2
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=180
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=182
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=183
+dilation=2
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=184
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=186
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=187
+dilation=2
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=188
+dilation=2
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+
+## Conv 5
+[convolutional]
+share_index=190
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=191
+dilation=2
+batch_normalize=1
+filters=512
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=192
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=194
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=195
+dilation=2
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=196
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=198
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=199
+dilation=2
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=200
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=1
+stride=1
+pad=1
+filters=1024
+activation=leaky
+
+[upsample]
+stride=2
+
+[route]
+layers = -1, 49
+
+[convolutional]
+batch_normalize=1
+size=1
+stride=1
+pad=1
+filters=1024
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=24
+activation=linear
+
+[yolo]
+mask = 4,5,6,7
+anchors = 8,8, 10,13, 16,30, 33,23,  32,32, 30,61, 62,45, 64,64,  59,119, 116,90, 156,198, 373,326
+classes=1
+num=12
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=0
+
+### TridentNet - medium objects - End
+
+
+
+
+
+
+
+
+
+
+
+### TridentNet - small objects - Start
+
+[route]
+layers = 165
+
+[convolutional]
+share_index=166
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=167
+dilation=1
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=168
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=170
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=171
+dilation=1
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=172
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=174
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=175
+dilation=1
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=176
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=178
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=179
+dilation=1
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=180
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=182
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=183
+dilation=1
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=184
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=186
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=187
+dilation=1
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=188
+dilation=1
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+
+## Conv 5
+[convolutional]
+share_index=190
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=191
+dilation=1
+batch_normalize=1
+filters=512
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=192
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=194
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=195
+dilation=1
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=196
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=198
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=199
+dilation=1
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=200
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=1
+stride=1
+pad=1
+filters=512
+activation=leaky
+
+[upsample]
+stride=4
+
+[route]
+layers = -1, 17
+
+[convolutional]
+batch_normalize=1
+size=1
+stride=1
+pad=1
+filters=512
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=24
+activation=linear
+
+[yolo]
+mask = 0,1,2,3
+anchors = 8,8, 10,13, 16,30, 33,23,  32,32, 30,61, 62,45, 64,64,  59,119, 116,90, 156,198, 373,326
+classes=1
+num=12
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=0
+
+### TridentNet - small objects - End
+
diff --git a/build/darknet/x64/partial.cmd b/build/darknet/x64/partial.cmd
index 03759e8a..5aa972f3 100644
--- a/build/darknet/x64/partial.cmd
+++ b/build/darknet/x64/partial.cmd
@@ -35,6 +35,8 @@ darknet.exe partial cfg/yolov3-tiny.cfg yolov3-tiny.weights yolov3-tiny.conv.15
 
 darknet.exe partial cfg/yolov3-tiny.cfg yolov3-tiny.weights yolov3-tiny.conv.14 14
 
+darknet.exe partial cfg/yolov3-tiny.cfg yolov3-tiny.weights yolov3-tiny.conv.13 13
+
 
 darknet.exe partial cfg/yolo9000.cfg yolo9000.weights yolo9000.conv.22 22
 
diff --git a/cfg/resnet152_trident.cfg b/cfg/resnet152_trident.cfg
new file mode 100644
index 00000000..c7e0e255
--- /dev/null
+++ b/cfg/resnet152_trident.cfg
@@ -0,0 +1,2177 @@
+[net]
+# Testing
+#batch=1
+#subdivisions=1
+# Training
+batch=64
+subdivisions=64
+width=608
+height=608
+channels=3
+momentum=0.9
+decay=0.0005
+angle=0
+saturation = 1.5
+exposure = 1.5
+hue=.1
+
+
+learning_rate=0.001
+burn_in=1000
+max_batches = 10000
+
+policy=sgdr
+sgdr_cycle=1000
+sgdr_mult=2
+steps=4000,6000,8000,9000
+#scales=1, 1, 0.1, 0.1
+
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=7
+stride=2
+pad=1
+activation=leaky
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+
+# Conv 4
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+
+
+
+### TridentNet - large objects - Start
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+
+## Conv 5
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=512
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+dilation=3
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=1
+stride=1
+pad=1
+filters=2048
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=24
+activation=linear
+
+[yolo]
+mask = 8,9,10,11
+anchors = 8,8, 10,13, 16,30, 33,23,  32,32, 30,61, 62,45, 59,119,   80,80, 116,90, 156,198, 373,326
+classes=1
+num=12
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=0
+
+### TridentNet - large objects - End
+
+
+
+
+
+
+
+### TridentNet - medium objects - Start
+
+[route]
+layers = 165
+
+[convolutional]
+share_index=166
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=167
+dilation=2
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=168
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=170
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=171
+dilation=2
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=172
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=174
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=175
+dilation=2
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=176
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=178
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=179
+dilation=2
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=180
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=182
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=183
+dilation=2
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=184
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=186
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=187
+dilation=2
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=188
+dilation=2
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+
+## Conv 5
+[convolutional]
+share_index=190
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=191
+dilation=2
+batch_normalize=1
+filters=512
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=192
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=194
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=195
+dilation=2
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=196
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=198
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=199
+dilation=2
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=200
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=1
+stride=1
+pad=1
+filters=1024
+activation=leaky
+
+[upsample]
+stride=2
+
+[route]
+layers = -1, 49
+
+[convolutional]
+batch_normalize=1
+size=1
+stride=1
+pad=1
+filters=1024
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=24
+activation=linear
+
+[yolo]
+mask = 4,5,6,7
+anchors = 8,8, 10,13, 16,30, 33,23,  32,32, 30,61, 62,45, 64,64,  59,119, 116,90, 156,198, 373,326
+classes=1
+num=12
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=0
+
+### TridentNet - medium objects - End
+
+
+
+
+
+
+
+
+
+
+
+### TridentNet - small objects - Start
+
+[route]
+layers = 165
+
+[convolutional]
+share_index=166
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=167
+dilation=1
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=168
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=170
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=171
+dilation=1
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=172
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=174
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=175
+dilation=1
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=176
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=178
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=179
+dilation=1
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=180
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=182
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=183
+dilation=1
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=184
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=186
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=187
+dilation=1
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=188
+dilation=1
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+
+## Conv 5
+[convolutional]
+share_index=190
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=191
+dilation=1
+batch_normalize=1
+filters=512
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=192
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=194
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=195
+dilation=1
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=196
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+share_index=198
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=199
+dilation=1
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+share_index=200
+batch_normalize=1
+filters=2048
+size=1
+stride=1
+pad=1
+activation=linear
+
+[shortcut]
+from=-4
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=1
+stride=1
+pad=1
+filters=512
+activation=leaky
+
+[upsample]
+stride=4
+
+[route]
+layers = -1, 17
+
+[convolutional]
+batch_normalize=1
+size=1
+stride=1
+pad=1
+filters=512
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=24
+activation=linear
+
+[yolo]
+mask = 0,1,2,3
+anchors = 8,8, 10,13, 16,30, 33,23,  32,32, 30,61, 62,45, 64,64,  59,119, 116,90, 156,198, 373,326
+classes=1
+num=12
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=0
+
+### TridentNet - small objects - End
+
diff --git a/include/darknet.h b/include/darknet.h
index 4154d664..e807ff9b 100644
--- a/include/darknet.h
+++ b/include/darknet.h
@@ -185,6 +185,7 @@ struct layer {
     void(*forward_gpu)   (struct layer, struct network_state);
     void(*backward_gpu)  (struct layer, struct network_state);
     void(*update_gpu)    (struct layer, int, float, float, float);
+    layer *share_layer;
     int batch_normalize;
     int shortcut;
     int batch;
diff --git a/src/coco.c b/src/coco.c
index c1535a35..cdfd3dff 100644
--- a/src/coco.c
+++ b/src/coco.c
@@ -218,7 +218,11 @@ void validate_coco(char *cfgfile, char *weightfile)
             free_image(val_resized[t]);
         }
     }
+#ifdef WIN32
+    fseek(fp, -3, SEEK_CUR);
+#else
     fseek(fp, -2, SEEK_CUR);
+#endif
     fprintf(fp, "\n]\n");
     fclose(fp);
 
diff --git a/src/conv_lstm_layer.c b/src/conv_lstm_layer.c
index 5caa3754..6cbaf1c3 100644
--- a/src/conv_lstm_layer.c
+++ b/src/conv_lstm_layer.c
@@ -66,44 +66,44 @@ layer make_conv_lstm_layer(int batch, int h, int w, int c, int output_filters, i
 
     // U
     l.uf = (layer*)calloc(1, sizeof(layer));
-    *(l.uf) = make_convolutional_layer(batch, steps, h, w, c, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+    *(l.uf) = make_convolutional_layer(batch, steps, h, w, c, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
     l.uf->batch = batch;
     if (l.workspace_size < l.uf->workspace_size) l.workspace_size = l.uf->workspace_size;
 
     l.ui = (layer*)calloc(1, sizeof(layer));
-    *(l.ui) = make_convolutional_layer(batch, steps, h, w, c, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+    *(l.ui) = make_convolutional_layer(batch, steps, h, w, c, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
     l.ui->batch = batch;
     if (l.workspace_size < l.ui->workspace_size) l.workspace_size = l.ui->workspace_size;
 
     l.ug = (layer*)calloc(1, sizeof(layer));
-    *(l.ug) = make_convolutional_layer(batch, steps, h, w, c, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+    *(l.ug) = make_convolutional_layer(batch, steps, h, w, c, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
     l.ug->batch = batch;
     if (l.workspace_size < l.ug->workspace_size) l.workspace_size = l.ug->workspace_size;
 
     l.uo = (layer*)calloc(1, sizeof(layer));
-    *(l.uo) = make_convolutional_layer(batch, steps, h, w, c, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+    *(l.uo) = make_convolutional_layer(batch, steps, h, w, c, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
     l.uo->batch = batch;
     if (l.workspace_size < l.uo->workspace_size) l.workspace_size = l.uo->workspace_size;
 
 
     // W
     l.wf = (layer*)calloc(1, sizeof(layer));
-    *(l.wf) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+    *(l.wf) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
     l.wf->batch = batch;
     if (l.workspace_size < l.wf->workspace_size) l.workspace_size = l.wf->workspace_size;
 
     l.wi = (layer*)calloc(1, sizeof(layer));
-    *(l.wi) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+    *(l.wi) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
     l.wi->batch = batch;
     if (l.workspace_size < l.wi->workspace_size) l.workspace_size = l.wi->workspace_size;
 
     l.wg = (layer*)calloc(1, sizeof(layer));
-    *(l.wg) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+    *(l.wg) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
     l.wg->batch = batch;
     if (l.workspace_size < l.wg->workspace_size) l.workspace_size = l.wg->workspace_size;
 
     l.wo = (layer*)calloc(1, sizeof(layer));
-    *(l.wo) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+    *(l.wo) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
     l.wo->batch = batch;
     if (l.workspace_size < l.wo->workspace_size) l.workspace_size = l.wo->workspace_size;
 
@@ -111,21 +111,21 @@ layer make_conv_lstm_layer(int batch, int h, int w, int c, int output_filters, i
     // V
     l.vf = (layer*)calloc(1, sizeof(layer));
     if (l.peephole) {
-        *(l.vf) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+        *(l.vf) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
         l.vf->batch = batch;
         if (l.workspace_size < l.vf->workspace_size) l.workspace_size = l.vf->workspace_size;
     }
 
     l.vi = (layer*)calloc(1, sizeof(layer));
     if (l.peephole) {
-        *(l.vi) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+        *(l.vi) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
         l.vi->batch = batch;
         if (l.workspace_size < l.vi->workspace_size) l.workspace_size = l.vi->workspace_size;
     }
 
     l.vo = (layer*)calloc(1, sizeof(layer));
     if (l.peephole) {
-        *(l.vo) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+        *(l.vo) = make_convolutional_layer(batch, steps, h, w, output_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
         l.vo->batch = batch;
         if (l.workspace_size < l.vo->workspace_size) l.workspace_size = l.vo->workspace_size;
     }
diff --git a/src/convolutional_layer.c b/src/convolutional_layer.c
index 84bc75fb..de77a585 100644
--- a/src/convolutional_layer.c
+++ b/src/convolutional_layer.c
@@ -274,10 +274,11 @@ void cudnn_convolutional_setup(layer *l, int cudnn_preference)
     CHECK_CUDNN(cudnnSetTensor4dDescriptor(l->normTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, 1, l->out_c, 1, 1));
     CHECK_CUDNN(cudnnSetTensor4dDescriptor(l->normDstTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, l->batch, l->out_c, l->out_h, l->out_w));
 
+    //printf("\n l->dilation = %d, l->pad = %d, l->size = %d \n", l->dilation, l->pad, l->size);
 #if(CUDNN_MAJOR >= 6)
-    CHECK_CUDNN(cudnnSetConvolution2dDescriptor(l->convDesc, l->pad, l->pad, l->stride, l->stride, l->dilation, l->dilation, CUDNN_CROSS_CORRELATION, CUDNN_DATA_FLOAT));    // cudnn >= 6.0
+    CHECK_CUDNN(cudnnSetConvolution2dDescriptor(l->convDesc, l->pad * l->dilation, l->pad* l->dilation, l->stride, l->stride, l->dilation, l->dilation, CUDNN_CROSS_CORRELATION, CUDNN_DATA_FLOAT));    // cudnn >= 6.0
 #else
-    CHECK_CUDNN(cudnnSetConvolution2dDescriptor(l->convDesc, l->pad, l->pad, l->stride, l->stride, l->dilation, l->dilation, CUDNN_CROSS_CORRELATION));    // cudnn 5.1
+    CHECK_CUDNN(cudnnSetConvolution2dDescriptor(l->convDesc, l->pad * l->dilation, l->pad * l->dilation, l->stride, l->stride, l->dilation, l->dilation, CUDNN_CROSS_CORRELATION));    // cudnn 5.1
 #endif
     int forward_algo = CUDNN_CONVOLUTION_FWD_PREFER_FASTEST;
     int backward_algo = CUDNN_CONVOLUTION_BWD_DATA_PREFER_FASTEST;
@@ -331,7 +332,7 @@ void cudnn_convolutional_setup(layer *l, int cudnn_preference)
 #endif
 #endif
 
-convolutional_layer make_convolutional_layer(int batch, int steps, int h, int w, int c, int n, int groups, int size, int stride, int dilation, int padding, ACTIVATION activation, int batch_normalize, int binary, int xnor, int adam, int use_bin_output, int index)
+convolutional_layer make_convolutional_layer(int batch, int steps, int h, int w, int c, int n, int groups, int size, int stride, int dilation, int padding, ACTIVATION activation, int batch_normalize, int binary, int xnor, int adam, int use_bin_output, int index, convolutional_layer *share_layer)
 {
     int total_batch = batch*steps;
     int i;
@@ -341,6 +342,7 @@ convolutional_layer make_convolutional_layer(int batch, int steps, int h, int w,
     if (xnor) groups = 1;   // disable groups for XNOR-net
     if (groups < 1) groups = 1;
 
+    l.share_layer = share_layer;
     l.index = index;
     l.h = h;
     l.w = w;
@@ -360,11 +362,20 @@ convolutional_layer make_convolutional_layer(int batch, int steps, int h, int w,
     l.learning_rate_scale = 1;
     l.nweights = (c / groups) * n * size * size;
 
-    l.weights = (float*)calloc(l.nweights, sizeof(float));
-    l.weight_updates = (float*)calloc(l.nweights, sizeof(float));
+    if (l.share_layer) {
+        l.weights = l.share_layer->weights;
+        l.weight_updates = l.share_layer->weight_updates;
 
-    l.biases = (float*)calloc(n, sizeof(float));
-    l.bias_updates = (float*)calloc(n, sizeof(float));
+        l.biases = l.share_layer->biases;
+        l.bias_updates = l.share_layer->bias_updates;
+    }
+    else {
+        l.weights = (float*)calloc(l.nweights, sizeof(float));
+        l.weight_updates = (float*)calloc(l.nweights, sizeof(float));
+
+        l.biases = (float*)calloc(n, sizeof(float));
+        l.bias_updates = (float*)calloc(n, sizeof(float));
+    }
 
     // float scale = 1./sqrt(size*size*c);
     float scale = sqrt(2./(size*size*c/groups));
@@ -411,20 +422,33 @@ convolutional_layer make_convolutional_layer(int batch, int steps, int h, int w,
     }
 
     if(batch_normalize){
-        l.scales = (float*)calloc(n, sizeof(float));
-        l.scale_updates = (float*)calloc(n, sizeof(float));
-        for(i = 0; i < n; ++i){
-            l.scales[i] = 1;
+        if (l.share_layer) {
+            l.scales = l.share_layer->scales;
+            l.scale_updates = l.share_layer->scale_updates;
+            l.mean = l.share_layer->mean;
+            l.variance = l.share_layer->variance;
+            l.mean_delta = l.share_layer->mean_delta;
+            l.variance_delta = l.share_layer->variance_delta;
+            l.rolling_mean = l.share_layer->rolling_mean;
+            l.rolling_variance = l.share_layer->rolling_variance;
+        }
+        else {
+            l.scales = (float*)calloc(n, sizeof(float));
+            l.scale_updates = (float*)calloc(n, sizeof(float));
+            for (i = 0; i < n; ++i) {
+                l.scales[i] = 1;
+            }
+
+            l.mean = (float*)calloc(n, sizeof(float));
+            l.variance = (float*)calloc(n, sizeof(float));
+
+            l.mean_delta = (float*)calloc(n, sizeof(float));
+            l.variance_delta = (float*)calloc(n, sizeof(float));
+
+            l.rolling_mean = (float*)calloc(n, sizeof(float));
+            l.rolling_variance = (float*)calloc(n, sizeof(float));
         }
 
-        l.mean = (float*)calloc(n, sizeof(float));
-        l.variance = (float*)calloc(n, sizeof(float));
-
-        l.mean_delta = (float*)calloc(n, sizeof(float));
-        l.variance_delta = (float*)calloc(n, sizeof(float));
-
-        l.rolling_mean = (float*)calloc(n, sizeof(float));
-        l.rolling_variance = (float*)calloc(n, sizeof(float));
         l.x = (float*)calloc(total_batch * l.outputs, sizeof(float));
         l.x_norm = (float*)calloc(total_batch * l.outputs, sizeof(float));
     }
@@ -452,16 +476,24 @@ convolutional_layer make_convolutional_layer(int batch, int steps, int h, int w,
             l.scale_m_gpu = cuda_make_array(l.scale_m, n);
             l.scale_v_gpu = cuda_make_array(l.scale_v, n);
         }
-
-        l.weights_gpu = cuda_make_array(l.weights, l.nweights);
-        l.weight_updates_gpu = cuda_make_array(l.weight_updates, l.nweights);
+        if (l.share_layer) {
+            l.weights_gpu = l.share_layer->weights_gpu;
+            l.weight_updates_gpu = l.share_layer->weight_updates_gpu;
+            l.weights_gpu16 = l.share_layer->weights_gpu16;
+            l.weight_updates_gpu16 = l.share_layer->weight_updates_gpu16;
+            l.biases_gpu = l.share_layer->biases_gpu;
+            l.bias_updates_gpu = l.share_layer->bias_updates_gpu;
+        }
+        else {
+            l.weights_gpu = cuda_make_array(l.weights, l.nweights);
+            l.weight_updates_gpu = cuda_make_array(l.weight_updates, l.nweights);
 #ifdef CUDNN_HALF
-        l.weights_gpu16 = cuda_make_array(NULL, l.nweights / 2 + 1);
-        l.weight_updates_gpu16 = cuda_make_array(NULL, l.nweights / 2 + 1);
+            l.weights_gpu16 = cuda_make_array(NULL, l.nweights / 2 + 1);
+            l.weight_updates_gpu16 = cuda_make_array(NULL, l.nweights / 2 + 1);
 #endif
-
-        l.biases_gpu = cuda_make_array(l.biases, n);
-        l.bias_updates_gpu = cuda_make_array(l.bias_updates, n);
+            l.biases_gpu = cuda_make_array(l.biases, n);
+            l.bias_updates_gpu = cuda_make_array(l.bias_updates, n);
+        }
 
         l.output_gpu = cuda_make_array(l.output, total_batch*out_h*out_w*n);
         l.delta_gpu = cuda_make_array(l.delta, total_batch*out_h*out_w*n);
@@ -476,18 +508,29 @@ convolutional_layer make_convolutional_layer(int batch, int steps, int h, int w,
         }
 
         if(batch_normalize){
-            l.scales_gpu = cuda_make_array(l.scales, n);
-            l.scale_updates_gpu = cuda_make_array(l.scale_updates, n);
+            if (l.share_layer) {
+                l.scales_gpu = l.share_layer->scales_gpu;
+                l.scale_updates_gpu = l.share_layer->scale_updates_gpu;
+                l.mean_gpu = l.share_layer->mean_gpu;
+                l.variance_gpu = l.share_layer->variance_gpu;
+                l.rolling_mean_gpu = l.share_layer->rolling_mean_gpu;
+                l.rolling_variance_gpu = l.share_layer->rolling_variance_gpu;
+                l.mean_delta_gpu = l.share_layer->mean_delta_gpu;
+                l.variance_delta_gpu = l.share_layer->variance_delta_gpu;
+            }
+            else {
+                l.scales_gpu = cuda_make_array(l.scales, n);
+                l.scale_updates_gpu = cuda_make_array(l.scale_updates, n);
 
-            l.mean_gpu = cuda_make_array(l.mean, n);
-            l.variance_gpu = cuda_make_array(l.variance, n);
+                l.mean_gpu = cuda_make_array(l.mean, n);
+                l.variance_gpu = cuda_make_array(l.variance, n);
 
-            l.rolling_mean_gpu = cuda_make_array(l.mean, n);
-            l.rolling_variance_gpu = cuda_make_array(l.variance, n);
-
-            l.mean_delta_gpu = cuda_make_array(l.mean, n);
-            l.variance_delta_gpu = cuda_make_array(l.variance, n);
+                l.rolling_mean_gpu = cuda_make_array(l.mean, n);
+                l.rolling_variance_gpu = cuda_make_array(l.variance, n);
 
+                l.mean_delta_gpu = cuda_make_array(l.mean, n);
+                l.variance_delta_gpu = cuda_make_array(l.variance, n);
+            }
             l.x_gpu = cuda_make_array(l.output, total_batch*out_h*out_w*n);
             l.x_norm_gpu = cuda_make_array(l.output, total_batch*out_h*out_w*n);
         }
@@ -504,7 +547,7 @@ convolutional_layer make_convolutional_layer(int batch, int steps, int h, int w,
     if (l.xnor && l.use_bin_output) fprintf(stderr, "convXB");
     else if (l.xnor) fprintf(stderr, "convX ");
     else fprintf(stderr, "conv  ");
-    fprintf(stderr, "%5d %2d x%2d /%2d  %4d x%4d x%4d   ->  %4d x%4d x%4d %5.3f BF\n", n, size, size, stride, w, h, c, l.out_w, l.out_h, l.out_c, l.bflops);
+    fprintf(stderr, "%5d %2d x%2d /%2d(%d)%4d x%4d x%4d   ->  %4d x%4d x%4d %5.3f BF\n", n, size, size, stride, dilation, w, h, c, l.out_w, l.out_h, l.out_c, l.bflops);
 
     return l;
 }
@@ -526,7 +569,7 @@ void denormalize_convolutional_layer(convolutional_layer l)
 
 void test_convolutional_layer()
 {
-    convolutional_layer l = make_convolutional_layer(1, 1, 5, 5, 3, 2, 1, 5, 2, 1, 1, LEAKY, 1, 0, 0, 0, 0, 0);
+    convolutional_layer l = make_convolutional_layer(1, 1, 5, 5, 3, 2, 1, 5, 2, 1, 1, LEAKY, 1, 0, 0, 0, 0, 0, NULL);
     l.batch_normalize = 1;
     float data[] = {1,1,1,1,1,
         1,1,1,1,1,
diff --git a/src/convolutional_layer.h b/src/convolutional_layer.h
index 586bd2cd..e937b4c5 100644
--- a/src/convolutional_layer.h
+++ b/src/convolutional_layer.h
@@ -30,7 +30,7 @@ void cuda_convert_f32_to_f16(float* input_f32, size_t size, float *output_f16);
 #endif
 
 size_t get_convolutional_workspace_size(layer l);
-convolutional_layer make_convolutional_layer(int batch, int steps, int h, int w, int c, int n, int groups, int size, int stride, int dilation, int padding, ACTIVATION activation, int batch_normalize, int binary, int xnor, int adam, int use_bin_output, int index);
+convolutional_layer make_convolutional_layer(int batch, int steps, int h, int w, int c, int n, int groups, int size, int stride, int dilation, int padding, ACTIVATION activation, int batch_normalize, int binary, int xnor, int adam, int use_bin_output, int index, convolutional_layer *share_layer);
 void denormalize_convolutional_layer(convolutional_layer l);
 void resize_convolutional_layer(convolutional_layer *layer, int w, int h);
 void forward_convolutional_layer(const convolutional_layer layer, network_state state);
diff --git a/src/crnn_layer.c b/src/crnn_layer.c
index d7c75b50..7609003b 100644
--- a/src/crnn_layer.c
+++ b/src/crnn_layer.c
@@ -50,17 +50,17 @@ layer make_crnn_layer(int batch, int h, int w, int c, int hidden_filters, int ou
     l.state = (float*)calloc(l.hidden * l.batch * (l.steps + 1), sizeof(float));
 
     l.input_layer = (layer*)calloc(1, sizeof(layer));
-    *(l.input_layer) = make_convolutional_layer(batch, steps, h, w, c, hidden_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+    *(l.input_layer) = make_convolutional_layer(batch, steps, h, w, c, hidden_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
     l.input_layer->batch = batch;
     if (l.workspace_size < l.input_layer->workspace_size) l.workspace_size = l.input_layer->workspace_size;
 
     l.self_layer = (layer*)calloc(1, sizeof(layer));
-    *(l.self_layer) = make_convolutional_layer(batch, steps, h, w, hidden_filters, hidden_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+    *(l.self_layer) = make_convolutional_layer(batch, steps, h, w, hidden_filters, hidden_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
     l.self_layer->batch = batch;
     if (l.workspace_size < l.self_layer->workspace_size) l.workspace_size = l.self_layer->workspace_size;
 
     l.output_layer = (layer*)calloc(1, sizeof(layer));
-    *(l.output_layer) = make_convolutional_layer(batch, steps, h, w, hidden_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0);
+    *(l.output_layer) = make_convolutional_layer(batch, steps, h, w, hidden_filters, output_filters, groups, size, stride, dilation, pad, activation, batch_normalize, 0, xnor, 0, 0, 0, NULL);
     l.output_layer->batch = batch;
     if (l.workspace_size < l.output_layer->workspace_size) l.workspace_size = l.output_layer->workspace_size;
 
diff --git a/src/layer.c b/src/layer.c
index 6409bd89..ade58f4e 100644
--- a/src/layer.c
+++ b/src/layer.c
@@ -66,12 +66,12 @@ void free_layer(layer l)
     if (l.concat)             free(l.concat);
     if (l.concat_delta)       free(l.concat_delta);
     if (l.binary_weights)     free(l.binary_weights);
-    if (l.biases)             free(l.biases);
-    if (l.bias_updates)       free(l.bias_updates);
-    if (l.scales)             free(l.scales);
-    if (l.scale_updates)      free(l.scale_updates);
-    if (l.weights)            free(l.weights);
-    if (l.weight_updates)     free(l.weight_updates);
+    if (l.biases)             free(l.biases), l.biases = NULL;
+    if (l.bias_updates)       free(l.bias_updates), l.bias_updates = NULL;
+    if (l.scales)             free(l.scales), l.scales = NULL;
+    if (l.scale_updates)      free(l.scale_updates), l.scale_updates = NULL;
+    if (l.weights)            free(l.weights), l.weights = NULL;
+    if (l.weight_updates)     free(l.weight_updates), l.weight_updates = NULL;
     if (l.align_bit_weights)  free(l.align_bit_weights);
     if (l.mean_arr)           free(l.mean_arr);
 #ifdef GPU
@@ -89,12 +89,12 @@ void free_layer(layer l)
     if (l.squared)            free(l.squared);
     if (l.norms)              free(l.norms);
     if (l.spatial_mean)       free(l.spatial_mean);
-    if (l.mean)               free(l.mean);
-    if (l.variance)           free(l.variance);
-    if (l.mean_delta)         free(l.mean_delta);
-    if (l.variance_delta)     free(l.variance_delta);
-    if (l.rolling_mean)       free(l.rolling_mean);
-    if (l.rolling_variance)   free(l.rolling_variance);
+    if (l.mean)               free(l.mean), l.mean = NULL;
+    if (l.variance)           free(l.variance), l.variance = NULL;
+    if (l.mean_delta)         free(l.mean_delta), l.mean_delta = NULL;
+    if (l.variance_delta)     free(l.variance_delta), l.variance_delta = NULL;
+    if (l.rolling_mean)       free(l.rolling_mean), l.rolling_mean = NULL;
+    if (l.rolling_variance)   free(l.rolling_variance), l.rolling_variance = NULL;
     if (l.x)                  free(l.x);
     if (l.x_norm)             free(l.x_norm);
     if (l.m)                  free(l.m);
@@ -143,12 +143,12 @@ void free_layer(layer l)
     if (l.concat_delta_gpu)        cuda_free(l.concat_delta_gpu);
     if (l.binary_input_gpu)        cuda_free(l.binary_input_gpu);
     if (l.binary_weights_gpu)      cuda_free(l.binary_weights_gpu);
-    if (l.mean_gpu)                cuda_free(l.mean_gpu);
-    if (l.variance_gpu)            cuda_free(l.variance_gpu);
-    if (l.rolling_mean_gpu)        cuda_free(l.rolling_mean_gpu);
-    if (l.rolling_variance_gpu)    cuda_free(l.rolling_variance_gpu);
-    if (l.variance_delta_gpu)      cuda_free(l.variance_delta_gpu);
-    if (l.mean_delta_gpu)          cuda_free(l.mean_delta_gpu);
+    if (l.mean_gpu)                cuda_free(l.mean_gpu), l.mean_gpu = NULL;
+    if (l.variance_gpu)            cuda_free(l.variance_gpu), l.variance_gpu = NULL;
+    if (l.rolling_mean_gpu)        cuda_free(l.rolling_mean_gpu), l.rolling_mean_gpu = NULL;
+    if (l.rolling_variance_gpu)    cuda_free(l.rolling_variance_gpu), l.rolling_variance_gpu = NULL;
+    if (l.variance_delta_gpu)      cuda_free(l.variance_delta_gpu), l.variance_delta_gpu = NULL;
+    if (l.mean_delta_gpu)          cuda_free(l.mean_delta_gpu), l.mean_delta_gpu = NULL;
     if (l.x_gpu)                   cuda_free(l.x_gpu);  // dont free
     if (l.x_norm_gpu)              cuda_free(l.x_norm_gpu);
 
@@ -157,14 +157,14 @@ void free_layer(layer l)
     if (l.align_workspace_gpu)     cuda_free(l.align_workspace_gpu);
     if (l.transposed_align_workspace_gpu) cuda_free(l.transposed_align_workspace_gpu);
 
-    if (l.weights_gpu)             cuda_free(l.weights_gpu);
-    if (l.weight_updates_gpu)      cuda_free(l.weight_updates_gpu);
-    if (l.weights_gpu16)           cuda_free(l.weights_gpu16);
-    if (l.weight_updates_gpu16)    cuda_free(l.weight_updates_gpu16);
-    if (l.biases_gpu)              cuda_free(l.biases_gpu);
-    if (l.bias_updates_gpu)        cuda_free(l.bias_updates_gpu);
-    if (l.scales_gpu)              cuda_free(l.scales_gpu);
-    if (l.scale_updates_gpu)       cuda_free(l.scale_updates_gpu);
+    if (l.weights_gpu)             cuda_free(l.weights_gpu), l.weights_gpu = NULL;
+    if (l.weight_updates_gpu)      cuda_free(l.weight_updates_gpu), l.weight_updates_gpu = NULL;
+    if (l.weights_gpu16)           cuda_free(l.weights_gpu16), l.weights_gpu16 = NULL;
+    if (l.weight_updates_gpu16)    cuda_free(l.weight_updates_gpu16), l.weight_updates_gpu16 = NULL;
+    if (l.biases_gpu)              cuda_free(l.biases_gpu), l.biases_gpu = NULL;
+    if (l.bias_updates_gpu)        cuda_free(l.bias_updates_gpu), l.bias_updates_gpu = NULL;
+    if (l.scales_gpu)              cuda_free(l.scales_gpu), l.scales_gpu = NULL;
+    if (l.scale_updates_gpu)       cuda_free(l.scale_updates_gpu), l.scale_updates_gpu = NULL;
     if (l.output_gpu)              cuda_free(l.output_gpu);
     if (l.delta_gpu)               cuda_free(l.delta_gpu);
     if (l.rand_gpu)                cuda_free(l.rand_gpu);
diff --git a/src/parser.c b/src/parser.c
index 0cb906a5..2c095132 100644
--- a/src/parser.c
+++ b/src/parser.c
@@ -147,13 +147,14 @@ local_layer parse_local(list *options, size_params params)
     return layer;
 }
 
-convolutional_layer parse_convolutional(list *options, size_params params)
+convolutional_layer parse_convolutional(list *options, size_params params, network net)
 {
     int n = option_find_int(options, "filters",1);
     int groups = option_find_int_quiet(options, "groups", 1);
     int size = option_find_int(options, "size",1);
     int stride = option_find_int(options, "stride",1);
     int dilation = option_find_int_quiet(options, "dilation", 1);
+    if (size == 1) dilation = 1;
     int pad = option_find_int_quiet(options, "pad",0);
     int padding = option_find_int_quiet(options, "padding",0);
     if(pad) padding = size/2;
@@ -161,6 +162,10 @@ convolutional_layer parse_convolutional(list *options, size_params params)
     char *activation_s = option_find_str(options, "activation", "logistic");
     ACTIVATION activation = get_activation(activation_s);
 
+    int share_index = option_find_int_quiet(options, "share_index", -1);
+    convolutional_layer *share_layer = NULL;
+    if(share_layer > -1) share_layer = &net.layers[share_index];
+
     int batch,h,w,c;
     h = params.h;
     w = params.w;
@@ -172,7 +177,7 @@ convolutional_layer parse_convolutional(list *options, size_params params)
     int xnor = option_find_int_quiet(options, "xnor", 0);
     int use_bin_output = option_find_int_quiet(options, "bin_output", 0);
 
-    convolutional_layer layer = make_convolutional_layer(batch,1,h,w,c,n,groups,size,stride,dilation,padding,activation, batch_normalize, binary, xnor, params.net.adam, use_bin_output, params.index);
+    convolutional_layer layer = make_convolutional_layer(batch,1,h,w,c,n,groups,size,stride,dilation,padding,activation, batch_normalize, binary, xnor, params.net.adam, use_bin_output, params.index, share_layer);
     layer.flipped = option_find_int_quiet(options, "flipped", 0);
     layer.dot = option_find_float_quiet(options, "dot", 0);
 
@@ -838,7 +843,7 @@ network parse_network_cfg_custom(char *filename, int batch, int time_steps)
         layer l = { (LAYER_TYPE)0 };
         LAYER_TYPE lt = string_to_layer_type(s->type);
         if(lt == CONVOLUTIONAL){
-            l = parse_convolutional(options, params);
+            l = parse_convolutional(options, params, net);
         }else if(lt == LOCAL){
             l = parse_local(options, params);
         }else if(lt == ACTIVE){
@@ -1112,7 +1117,7 @@ void save_weights_upto(network net, char *filename, int cutoff)
     int i;
     for(i = 0; i < net.n && i < cutoff; ++i){
         layer l = net.layers[i];
-        if(l.type == CONVOLUTIONAL){
+        if(l.type == CONVOLUTIONAL && l.share_layer == NULL){
             save_convolutional_weights(l, fp);
         } if(l.type == CONNECTED){
             save_connected_weights(l, fp);
@@ -1337,7 +1342,7 @@ void load_weights_upto(network *net, char *filename, int cutoff)
     for(i = 0; i < net->n && i < cutoff; ++i){
         layer l = net->layers[i];
         if (l.dontload) continue;
-        if(l.type == CONVOLUTIONAL){
+        if(l.type == CONVOLUTIONAL && l.share_layer == NULL){
             load_convolutional_weights(l, fp);
         }
         if(l.type == CONNECTED){