updates and things

.gitignore

@@ -14,6 +14,7 @@ decaf/
 submission/
 cfg/
 darknet
+.fuse*
 
 # OS Generated #
 .DS_Store*

Makefile

@@ -1,6 +1,6 @@
-GPU=1
-CUDNN=1
-OPENCV=1
+GPU=0
+CUDNN=0
+OPENCV=0
 DEBUG=0
 
 ARCH= --gpu-architecture=compute_52 --gpu-code=compute_52
@@ -41,7 +41,7 @@ CFLAGS+= -DCUDNN
 LDFLAGS+= -lcudnn
 endif
 
-OBJ=gemm.o utils.o cuda.o deconvolutional_layer.o convolutional_layer.o list.o image.o activations.o im2col.o col2im.o blas.o crop_layer.o dropout_layer.o maxpool_layer.o softmax_layer.o data.o matrix.o network.o connected_layer.o cost_layer.o parser.o option_list.o darknet.o detection_layer.o imagenet.o captcha.o route_layer.o writing.o box.o nightmare.o normalization_layer.o avgpool_layer.o coco.o dice.o yolo.o detector.o layer.o compare.o classifier.o local_layer.o swag.o shortcut_layer.o activation_layer.o rnn_layer.o gru_layer.o rnn.o rnn_vid.o crnn_layer.o demo.o tag.o cifar.o go.o batchnorm_layer.o art.o region_layer.o reorg_layer.o super.o voxel.o
+OBJ=gemm.o utils.o cuda.o deconvolutional_layer.o convolutional_layer.o list.o image.o activations.o im2col.o col2im.o blas.o crop_layer.o dropout_layer.o maxpool_layer.o softmax_layer.o data.o matrix.o network.o connected_layer.o cost_layer.o parser.o option_list.o darknet.o detection_layer.o captcha.o route_layer.o writing.o box.o nightmare.o normalization_layer.o avgpool_layer.o coco.o dice.o yolo.o detector.o layer.o compare.o classifier.o local_layer.o swag.o shortcut_layer.o activation_layer.o rnn_layer.o gru_layer.o rnn.o rnn_vid.o crnn_layer.o demo.o tag.o cifar.o go.o batchnorm_layer.o art.o region_layer.o reorg_layer.o super.o voxel.o
 ifeq ($(GPU), 1) 
 LDFLAGS+= -lstdc++ 
 OBJ+=convolutional_kernels.o deconvolutional_kernels.o activation_kernels.o im2col_kernels.o col2im_kernels.o blas_kernels.o crop_layer_kernels.o dropout_layer_kernels.o maxpool_layer_kernels.o softmax_layer_kernels.o network_kernels.o avgpool_layer_kernels.o

cfg/extraction22k.cfg

@@ -0,0 +1,209 @@
+[net]
+batch=128
+subdivisions=1
+height=224
+width=224
+max_crop=320
+channels=3
+momentum=0.9
+decay=0.0005
+
+learning_rate=0.01
+max_batches = 0
+policy=steps
+steps=444000,590000,970000
+scales=.5,.2,.1
+
+#policy=sigmoid
+#gamma=.00008
+#step=100000
+#max_batches=200000
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=7
+stride=2
+pad=1
+activation=leaky
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=192
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=128
+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=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=2048
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=2048
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[avgpool]
+
+[connected]
+output=21842
+activation=leaky
+
+[softmax]
+groups=1
+
+[cost]
+type=sse
+

cfg/go.test.cfg

@@ -3,102 +3,126 @@ batch=1
 subdivisions=1
 height=19
 width=19
-channels=8
+channels=1
 momentum=0.9
 decay=0.0005
 
 learning_rate=0.1
-max_batches = 0
-policy=steps
-steps=50000
-scales=.1
+policy=poly
+power=4
+max_batches=400000
 
 [convolutional]
-filters=512
+filters=192
 size=3
 stride=1
 pad=1
-activation=leaky
+activation=relu
 batch_normalize=1
 
 [convolutional]
-filters=256
-size=1
-stride=1
-pad=1
-activation=leaky
-batch_normalize=1
-
-[convolutional]
-filters=512
+filters=192
 size=3
 stride=1
 pad=1
-activation=leaky
+activation=relu
 batch_normalize=1
 
 [convolutional]
-filters=256
-size=1
-stride=1
-pad=1
-activation=leaky
-batch_normalize=1
-
-[convolutional]
-filters=512
+filters=192
 size=3
 stride=1
 pad=1
-activation=leaky
+activation=relu
 batch_normalize=1
 
 [convolutional]
-filters=256
-size=1
-stride=1
-pad=1
-activation=leaky
-batch_normalize=1
-
-[convolutional]
-filters=512
+filters=192
 size=3
 stride=1
 pad=1
-activation=leaky
+activation=relu
 batch_normalize=1
 
 [convolutional]
-filters=256
-size=1
-stride=1
-pad=1
-activation=leaky
-batch_normalize=1
-
-[convolutional]
-filters=512
+filters=192
 size=3
 stride=1
 pad=1
-activation=leaky
+activation=relu
 batch_normalize=1
 
 [convolutional]
-filters=256
-size=1
+filters=192
+size=3
 stride=1
 pad=1
-activation=leaky
+activation=relu
 batch_normalize=1
 
+[convolutional]
+filters=192
+size=3
+stride=1
+pad=1
+activation=relu
+batch_normalize=1
+
+[convolutional]
+filters=192
+size=3
+stride=1
+pad=1
+activation=relu
+batch_normalize=1
+
+[convolutional]
+filters=192
+size=3
+stride=1
+pad=1
+activation=relu
+batch_normalize=1
+
+[convolutional]
+filters=192
+size=3
+stride=1
+pad=1
+activation=relu
+batch_normalize=1
+
+[convolutional]
+filters=192
+size=3
+stride=1
+pad=1
+activation=relu
+batch_normalize=1
+
+[convolutional]
+filters=192
+size=3
+stride=1
+pad=1
+activation=relu
+batch_normalize=1
+
+[convolutional]
+filters=192
+size=3
+stride=1
+pad=1
+activation=relu
+batch_normalize=1
+
+
 [convolutional]
 filters=1
 size=1
 stride=1
 pad=1
-activation=leaky
+activation=linear
 
 [softmax]
 

cfg/imagenet1k.dataset

@@ -1,9 +1,8 @@
 classes=1000
-labels = data/inet.labels.list
-names = data/shortnames.txt
 train  = /data/imagenet/imagenet1k.train.list
 valid  = /data/imagenet/imagenet1k.valid.list
-top=5
-test = /Users/pjreddie/Documents/sites/selfie/paths.list
 backup = /home/pjreddie/backup/
+labels = data/imagenet.labels.list
+names  = data/imagenet.shortnames.list
+top=5
 

cfg/yolo.cfg

@@ -1,11 +1,14 @@
 [net]
-batch=64
-subdivisions=2
+batch=1
+subdivisions=1
 height=448
 width=448
 channels=3
 momentum=0.9
 decay=0.0005
+saturation=1.5
+exposure=1.5
+hue=.1
 
 learning_rate=0.0005
 policy=steps
@@ -13,15 +16,6 @@ steps=200,400,600,20000,30000
 scales=2.5,2,2,.1,.1
 max_batches = 40000
 
-[crop]
-crop_width=448
-crop_height=448
-flip=0
-angle=0
-saturation = 1.5
-exposure = 1.5
-noadjust=1
-
 [convolutional]
 batch_normalize=1
 filters=64

cfg/yolo.train.cfg

@@ -0,0 +1,257 @@
+[net]
+batch=64
+subdivisions=4
+height=448
+width=448
+channels=3
+momentum=0.9
+decay=0.0005
+saturation=1.5
+exposure=1.5
+hue=.1
+
+learning_rate=0.0005
+policy=steps
+steps=200,400,600,20000,30000
+scales=2.5,2,2,.1,.1
+max_batches = 40000
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=7
+stride=2
+pad=1
+activation=leaky
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=192
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=128
+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=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=3
+stride=1
+pad=1
+activation=leaky
+
+#######
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=1024
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=2
+pad=1
+filters=1024
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=1024
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=1024
+activation=leaky
+
+[local]
+size=3
+stride=1
+pad=1
+filters=256
+activation=leaky
+
+[dropout]
+probability=.5
+
+[connected]
+output= 1715
+activation=linear
+
+[detection]
+classes=20
+coords=4
+rescore=1
+side=7
+num=3
+softmax=0
+sqrt=1
+jitter=.2
+
+object_scale=1
+noobject_scale=.5
+class_scale=1
+coord_scale=5
+

src/blas.h

@@ -40,6 +40,7 @@ void axpy_ongpu_offset(int N, float ALPHA, float * X, int OFFX, int INCX, float
 void copy_ongpu(int N, float * X, int INCX, float * Y, int INCY);
 void copy_ongpu_offset(int N, float * X, int OFFX, int INCX, float * Y, int OFFY, int INCY);
 void scal_ongpu(int N, float ALPHA, float * X, int INCX);
+void supp_ongpu(int N, float ALPHA, float * X, int INCX);
 void mask_ongpu(int N, float * X, float mask_num, float * mask);
 void const_ongpu(int N, float ALPHA, float *X, int INCX);
 void pow_ongpu(int N, float ALPHA, float *X, int INCX, float *Y, int INCY);

src/blas_kernels.cu

@@ -368,6 +368,14 @@ __global__ void constrain_kernel(int N, float ALPHA, float *X, int INCX)
     if(i < N) X[i*INCX] = min(ALPHA, max(-ALPHA, X[i*INCX]));
 }
 
+__global__ void supp_kernel(int N, float ALPHA, float *X, int INCX)
+{
+    int i = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
+    if(i < N) {
+        if((X[i*INCX] * X[i*INCX]) < (ALPHA * ALPHA)) X[i*INCX] = 0;
+    }
+}
+
 __global__ void scal_kernel(int N, float ALPHA, float *X, int INCX)
 {
     int i = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
@@ -552,6 +560,12 @@ extern "C" void scal_ongpu(int N, float ALPHA, float * X, int INCX)
     check_error(cudaPeekAtLastError());
 }
 
+extern "C" void supp_ongpu(int N, float ALPHA, float * X, int INCX)
+{
+    supp_kernel<<<cuda_gridsize(N), BLOCK>>>(N, ALPHA, X, INCX);
+    check_error(cudaPeekAtLastError());
+}
+
 extern "C" void fill_ongpu(int N, float ALPHA, float * X, int INCX)
 {
     fill_kernel<<<cuda_gridsize(N), BLOCK>>>(N, ALPHA, X, INCX);
@@ -633,6 +647,7 @@ extern "C" void l2_gpu(int n, float *pred, float *truth, float *delta, float *er
 }
 
 
+
 __global__ void weighted_sum_kernel(int n, float *a, float *b, float *s, float *c)
 {
     int i = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;

src/classifier.c

@@ -39,6 +39,18 @@ list *read_data_cfg(char *filename)
     return options;
 }
 
+float *get_regression_values(char **labels, int n)
+{
+    float *v = calloc(n, sizeof(float));
+    int i;
+    for(i = 0; i < n; ++i){
+        char *p = strchr(labels[i], ' ');
+        *p = 0;
+        v[i] = atof(p+1);
+    }
+    return v;
+}
+
 void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int clear)
 {
     int nthreads = 8;
@@ -85,6 +97,7 @@ void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int clear)
     args.angle = net.angle;
     args.exposure = net.exposure;
     args.saturation = net.saturation;
+    args.hue = net.hue;
     args.size = net.w;
 
     args.paths = paths;
@@ -116,6 +129,7 @@ void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int clear)
         printf("Loaded: %lf seconds\n", sec(clock()-time));
         time=clock();
 
+        #ifdef OPENCV
         if(0){
             int u;
             for(u = 0; u < imgs; ++u){
@@ -124,6 +138,7 @@ void train_classifier(char *datacfg, char *cfgfile, char *weightfile, int clear)
                 cvWaitKey(0);
             }
         }
+        #endif
 
         float loss = train_network(net, train);
         if(avg_loss == -1) avg_loss = loss;
@@ -440,7 +455,7 @@ void validate_classifier_multi(char *datacfg, char *filename, char *weightfile)
 
     char **labels = get_labels(label_list);
     list *plist = get_paths(valid_list);
-    int scales[] = {192, 224, 288, 320, 352};
+    int scales[] = {224, 288, 320, 352, 384};
     int nscales = sizeof(scales)/sizeof(scales[0]);
 
     char **paths = (char **)list_to_array(plist);
@@ -484,6 +499,88 @@ void validate_classifier_multi(char *datacfg, char *filename, char *weightfile)
     }
 }
 
+void try_classifier(char *datacfg, char *cfgfile, char *weightfile, char *filename, int layer_num)
+{
+    network net = parse_network_cfg(cfgfile);
+    if(weightfile){
+        load_weights(&net, weightfile);
+    }
+    set_batch_network(&net, 1);
+    srand(2222222);
+
+    list *options = read_data_cfg(datacfg);
+
+    char *name_list = option_find_str(options, "names", 0);
+    if(!name_list) name_list = option_find_str(options, "labels", "data/labels.list");
+    int top = option_find_int(options, "top", 1);
+
+    int i = 0;
+    char **names = get_labels(name_list);
+    clock_t time;
+    int *indexes = calloc(top, sizeof(int));
+    char buff[256];
+    char *input = buff;
+    while(1){
+        if(filename){
+            strncpy(input, filename, 256);
+        }else{
+            printf("Enter Image Path: ");
+            fflush(stdout);
+            input = fgets(input, 256, stdin);
+            if(!input) return;
+            strtok(input, "\n");
+        }
+        image orig = load_image_color(input, 0, 0);
+        image r = resize_min(orig, 256);
+        image im = crop_image(r, (r.w - 224 - 1)/2 + 1, (r.h - 224 - 1)/2 + 1, 224, 224);
+        float mean[] = {0.48263312050943, 0.45230225481413, 0.40099074308742};
+        float std[] = {0.22590347483426, 0.22120921437787, 0.22103996251583};
+        float var[3];
+        var[0] = std[0]*std[0];
+        var[1] = std[1]*std[1];
+        var[2] = std[2]*std[2];
+
+        normalize_cpu(im.data, mean, var, 1, 3, im.w*im.h);
+
+        float *X = im.data;
+        time=clock();
+        float *predictions = network_predict(net, X);
+        
+        layer l = net.layers[layer_num];
+        for(i = 0; i < l.c; ++i){
+        if(l.rolling_mean) printf("%f %f %f\n", l.rolling_mean[i], l.rolling_variance[i], l.scales[i]);
+        }
+        #ifdef GPU
+        cuda_pull_array(l.output_gpu, l.output, l.outputs);
+        #endif
+        for(i = 0; i < l.outputs; ++i){
+            printf("%f\n", l.output[i]);
+        }
+        /*
+        
+        printf("\n\nWeights\n");
+        for(i = 0; i < l.n*l.size*l.size*l.c; ++i){
+            printf("%f\n", l.filters[i]);
+        }
+
+        printf("\n\nBiases\n");
+        for(i = 0; i < l.n; ++i){
+            printf("%f\n", l.biases[i]);
+        }
+        */
+
+        top_predictions(net, top, indexes);
+        printf("%s: Predicted in %f seconds.\n", input, sec(clock()-time));
+        for(i = 0; i < top; ++i){
+            int index = indexes[i];
+            printf("%s: %f\n", names[index], predictions[index]);
+        }
+        free_image(im);
+        if (filename) break;
+    }
+}
+
+
 void predict_classifier(char *datacfg, char *cfgfile, char *weightfile, char *filename)
 {
     network net = parse_network_cfg(cfgfile);
@@ -649,6 +746,127 @@ void test_classifier(char *datacfg, char *cfgfile, char *weightfile, int target_
 }
 
 
+void threat_classifier(char *datacfg, char *cfgfile, char *weightfile, int cam_index, const char *filename)
+{
+#ifdef OPENCV
+    float threat = 0;
+    float roll = .2;
+
+    printf("Classifier Demo\n");
+    network net = parse_network_cfg(cfgfile);
+    if(weightfile){
+        load_weights(&net, weightfile);
+    }
+    set_batch_network(&net, 1);
+    list *options = read_data_cfg(datacfg);
+
+    srand(2222222);
+    CvCapture * cap;
+
+    if(filename){
+        cap = cvCaptureFromFile(filename);
+    }else{
+        cap = cvCaptureFromCAM(cam_index);
+    }
+
+    int top = option_find_int(options, "top", 1);
+
+    char *name_list = option_find_str(options, "names", 0);
+    char **names = get_labels(name_list);
+
+    int *indexes = calloc(top, sizeof(int));
+
+    if(!cap) error("Couldn't connect to webcam.\n");
+    //cvNamedWindow("Threat", CV_WINDOW_NORMAL); 
+    //cvResizeWindow("Threat", 512, 512);
+    float fps = 0;
+    int i;
+
+    int count = 0;
+
+    while(1){
+        ++count;
+        struct timeval tval_before, tval_after, tval_result;
+        gettimeofday(&tval_before, NULL);
+
+        image in = get_image_from_stream(cap);
+        if(!in.data) break;
+        image in_s = resize_image(in, net.w, net.h);
+
+    image out = in;
+    int x1 = out.w / 20;
+    int y1 = out.h / 20;
+    int x2 = 2*x1;
+    int y2 = out.h - out.h/20;
+
+    int border = .01*out.h;
+    int h = y2 - y1 - 2*border;
+    int w = x2 - x1 - 2*border;
+
+        float *predictions = network_predict(net, in_s.data);
+        float curr_threat = predictions[0] * 0 + predictions[1] * .6 + predictions[2];
+        threat = roll * curr_threat + (1-roll) * threat;
+
+        draw_box_width(out, x2 + border, y1 + .02*h, x2 + .5 * w, y1 + .02*h + border, border, 0,0,0);
+        if(threat > .97) {
+            draw_box_width(out,  x2 + .5 * w + border,
+                    y1 + .02*h - 2*border, 
+                    x2 + .5 * w + 6*border, 
+                    y1 + .02*h + 3*border, 3*border, 1,0,0);
+        }
+        draw_box_width(out,  x2 + .5 * w + border,
+                y1 + .02*h - 2*border, 
+                x2 + .5 * w + 6*border, 
+                y1 + .02*h + 3*border, .5*border, 0,0,0);
+        draw_box_width(out, x2 + border, y1 + .42*h, x2 + .5 * w, y1 + .42*h + border, border, 0,0,0);
+        if(threat > .57) {
+        draw_box_width(out,  x2 + .5 * w + border,
+                y1 + .42*h - 2*border, 
+                x2 + .5 * w + 6*border, 
+                y1 + .42*h + 3*border, 3*border, 1,1,0);
+            }
+        draw_box_width(out,  x2 + .5 * w + border,
+                y1 + .42*h - 2*border, 
+                x2 + .5 * w + 6*border, 
+                y1 + .42*h + 3*border, .5*border, 0,0,0);
+
+        draw_box_width(out, x1, y1, x2, y2, border, 0,0,0);
+        for(i = 0; i < threat * h ; ++i){
+            float ratio = (float) i / h;
+            float r = (ratio < .5) ? (2*(ratio)) : 1;
+            float g = (ratio < .5) ? 1 : 1 - 2*(ratio - .5);
+            draw_box_width(out, x1 + border, y2 - border - i, x2 - border, y2 - border - i, 1, r, g, 0);
+        }
+        top_predictions(net, top, indexes);
+        char buff[256];
+        sprintf(buff, "/home/pjreddie/tmp/threat_%06d", count);
+        save_image(out, buff);
+
+        printf("\033[2J");
+        printf("\033[1;1H");
+        printf("\nFPS:%.0f\n",fps);
+
+        for(i = 0; i < top; ++i){
+            int index = indexes[i];
+            printf("%.1f%%: %s\n", predictions[index]*100, names[index]);
+        }
+
+        if(0){
+            show_image(out, "Threat");
+            cvWaitKey(10);
+        }
+        free_image(in_s);
+        free_image(in);
+
+        gettimeofday(&tval_after, NULL);
+        timersub(&tval_after, &tval_before, &tval_result);
+        float curr = 1000000.f/((long int)tval_result.tv_usec);
+        fps = .9*fps + .1*curr;
+    }
+#endif
+}
+
+
 void demo_classifier(char *datacfg, char *cfgfile, char *weightfile, int cam_index, const char *filename)
 {
 #ifdef OPENCV
@@ -732,8 +950,10 @@ void run_classifier(int argc, char **argv)
     char *layer_s = (argc > 7) ? argv[7]: 0;
     int layer = layer_s ? atoi(layer_s) : -1;
     if(0==strcmp(argv[2], "predict")) predict_classifier(data, cfg, weights, filename);
+    else if(0==strcmp(argv[2], "try")) try_classifier(data, cfg, weights, filename, atoi(layer_s));
     else if(0==strcmp(argv[2], "train")) train_classifier(data, cfg, weights, clear);
     else if(0==strcmp(argv[2], "demo")) demo_classifier(data, cfg, weights, cam_index, filename);
+    else if(0==strcmp(argv[2], "threat")) threat_classifier(data, cfg, weights, cam_index, filename);
     else if(0==strcmp(argv[2], "test")) test_classifier(data, cfg, weights, layer);
     else if(0==strcmp(argv[2], "label")) label_classifier(data, cfg, weights);
     else if(0==strcmp(argv[2], "valid")) validate_classifier_single(data, cfg, weights);

src/coco.c

@@ -25,6 +25,7 @@ void train_coco(char *cfgfile, char *weightfile)
     //char *train_images = "/home/pjreddie/data/voc/test/train.txt";
     //char *train_images = "/home/pjreddie/data/coco/train.txt";
     char *train_images = "data/coco.trainval.txt";
+    //char *train_images = "data/bags.train.list";
     char *backup_directory = "/home/pjreddie/backup/";
     srand(time(0));
     data_seed = time(0);
@@ -63,6 +64,11 @@ void train_coco(char *cfgfile, char *weightfile)
     args.d = &buffer;
     args.type = REGION_DATA;
 
+    args.angle = net.angle;
+    args.exposure = net.exposure;
+    args.saturation = net.saturation;
+    args.hue = net.hue;
+
     pthread_t load_thread = load_data_in_thread(args);
     clock_t time;
     //while(i*imgs < N*120){
@@ -94,6 +100,11 @@ void train_coco(char *cfgfile, char *weightfile)
             sprintf(buff, "%s/%s_%d.weights", backup_directory, base, i);
             save_weights(net, buff);
         }
+        if(i%100==0){
+            char buff[256];
+            sprintf(buff, "%s/%s.backup", backup_directory, base);
+            save_weights(net, buff);
+        }
         free_data(train);
     }
     char buff[256];

src/col2im.c

@@ -16,13 +16,9 @@ void col2im_cpu(float* data_col,
          int ksize,  int stride, int pad, float* data_im) 
 {
     int c,h,w;
-    int height_col = (height - ksize) / stride + 1;
-    int width_col = (width - ksize) / stride + 1;
-    if (pad){
-        height_col = 1 + (height-1) / stride;
-        width_col = 1 + (width-1) / stride;
-        pad = ksize/2;
-    }
+    int height_col = (height + 2*pad - ksize) / stride + 1;
+    int width_col = (width + 2*pad - ksize) / stride + 1;
+
     int channels_col = channels * ksize * ksize;
     for (c = 0; c < channels_col; ++c) {
         int w_offset = c % ksize;

src/col2im_kernels.cu

@@ -46,7 +46,6 @@ void col2im_ongpu(float *data_col,
         int ksize, int stride, int pad, float *data_im){
     // We are going to launch channels * height_col * width_col kernels, each
     // kernel responsible for copying a single-channel grid.
-    pad = pad ? ksize/2 : 0;
     int height_col = (height + 2 * pad - ksize) / stride + 1;
     int width_col = (width + 2 * pad - ksize) / stride + 1;
     int num_kernels = channels * height * width;

src/convolutional_kernels.cu

@@ -17,7 +17,7 @@ __global__ void binarize_kernel(float *x, int n, float *binary)
 {
     int i = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
     if (i >= n) return;
-    binary[i] = (x[i] > 0) ? 1 : -1;
+    binary[i] = (x[i] >= 0) ? 1 : -1;
 }
 
 void binarize_gpu(float *x, int n, float *binary)
@@ -60,6 +60,7 @@ __global__ void binarize_filters_kernel(float *filters, int n, int size, float *
     mean = mean / size;
     for(i = 0; i < size; ++i){
         binary[f*size + i] = (filters[f*size + i] > 0) ? mean : -mean;
+        //binary[f*size + i] = filters[f*size + i];
     }
 }
 

src/convolutional_layer.c

@@ -70,18 +70,12 @@ void binarize_input(float *input, int n, int size, float *binary)
 
 int convolutional_out_height(convolutional_layer l)
 {
-    int h = l.h;
-    if (!l.pad) h -= l.size;
-    else h -= 1;
-    return h/l.stride + 1;
+    return (l.h + 2*l.pad - l.size) / l.stride + 1;
 }
 
 int convolutional_out_width(convolutional_layer l)
 {
-    int w = l.w;
-    if (!l.pad) w -= l.size;
-    else w -= 1;
-    return w/l.stride + 1;
+    return (l.w + 2*l.pad - l.size) / l.stride + 1;
 }
 
 image get_convolutional_image(convolutional_layer l)
@@ -148,8 +142,7 @@ void cudnn_convolutional_setup(layer *l)
     cudnnSetTensor4dDescriptor(l->srcTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, l->batch, l->c, l->h, l->w); 
     cudnnSetTensor4dDescriptor(l->dstTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, l->batch, l->out_c, l->out_h, l->out_w); 
     cudnnSetFilter4dDescriptor(l->filterDesc, CUDNN_DATA_FLOAT, CUDNN_TENSOR_NCHW, l->n, l->c, l->size, l->size); 
-    int padding = l->pad ? l->size/2 : 0;
-    cudnnSetConvolution2dDescriptor(l->convDesc, padding, padding, l->stride, l->stride, 1, 1, CUDNN_CROSS_CORRELATION);
+    cudnnSetConvolution2dDescriptor(l->convDesc, l->pad, l->pad, l->stride, l->stride, 1, 1, CUDNN_CROSS_CORRELATION);
     cudnnGetConvolutionForwardAlgorithm(cudnn_handle(),
             l->srcTensorDesc,
             l->filterDesc,
@@ -178,7 +171,7 @@ void cudnn_convolutional_setup(layer *l)
 #endif
 #endif
 
-convolutional_layer make_convolutional_layer(int batch, int h, int w, int c, int n, int size, int stride, int pad, ACTIVATION activation, int batch_normalize, int binary, int xnor)
+convolutional_layer make_convolutional_layer(int batch, int h, int w, int c, int n, int size, int stride, int padding, ACTIVATION activation, int batch_normalize, int binary, int xnor)
 {
     int i;
     convolutional_layer l = {0};
@@ -193,7 +186,7 @@ convolutional_layer make_convolutional_layer(int batch, int h, int w, int c, int
     l.batch = batch;
     l.stride = stride;
     l.size = size;
-    l.pad = pad;
+    l.pad = padding;
     l.batch_normalize = batch_normalize;
 
     l.filters = calloc(c*n*size*size, sizeof(float));

src/cost_layer.c

@@ -98,6 +98,15 @@ void push_cost_layer(cost_layer l)
     cuda_push_array(l.delta_gpu, l.delta, l.batch*l.inputs);
 }
 
+int float_abs_compare (const void * a, const void * b)
+{
+    float fa = *(const float*) a;
+    if(fa < 0) fa = -fa;
+    float fb = *(const float*) b;
+    if(fb < 0) fb = -fb;
+    return (fa > fb) - (fa < fb);
+}
+
 void forward_cost_layer_gpu(cost_layer l, network_state state)
 {
     if (!state.truth) return;
@@ -111,6 +120,16 @@ void forward_cost_layer_gpu(cost_layer l, network_state state)
         l2_gpu(l.batch*l.inputs, state.input, state.truth, l.delta_gpu, l.output_gpu);
     }
 
+    if(l.ratio){
+        cuda_pull_array(l.delta_gpu, l.delta, l.batch*l.inputs);
+        qsort(l.delta, l.batch*l.inputs, sizeof(float), float_abs_compare);
+        int n = (1-l.ratio) * l.batch*l.inputs;
+        float thresh = l.delta[n];
+        thresh = 0;
+        printf("%f\n", thresh);
+        supp_ongpu(l.batch*l.inputs, thresh, l.delta_gpu, 1);
+    }
+
     cuda_pull_array(l.output_gpu, l.output, l.batch*l.inputs);
     l.cost[0] = sum_array(l.output, l.batch*l.inputs);
 }

src/darknet.c

@@ -13,7 +13,6 @@
 #endif
 
 extern void run_voxel(int argc, char **argv);
-extern void run_imagenet(int argc, char **argv);
 extern void run_yolo(int argc, char **argv);
 extern void run_detector(int argc, char **argv);
 extern void run_coco(int argc, char **argv);
@@ -327,9 +326,7 @@ int main(int argc, char **argv)
     }
 #endif
 
-    if(0==strcmp(argv[1], "imagenet")){
-        run_imagenet(argc, argv);
-    } else if (0 == strcmp(argv[1], "average")){
+    if (0 == strcmp(argv[1], "average")){
         average(argc, argv);
     } else if (0 == strcmp(argv[1], "yolo")){
         run_yolo(argc, argv);

src/data.c

@@ -100,7 +100,7 @@ matrix load_image_paths(char **paths, int n, int w, int h)
     return X;
 }
 
-matrix load_image_augment_paths(char **paths, int n, int min, int max, int size, float angle, float exposure, float saturation)
+matrix load_image_augment_paths(char **paths, int n, int min, int max, int size, float angle, float hue, float saturation, float exposure)
 {
     int i;
     matrix X;
@@ -113,10 +113,7 @@ matrix load_image_augment_paths(char **paths, int n, int min, int max, int size,
         image crop = random_augment_image(im, angle, min, max, size);
         int flip = rand_r(&data_seed)%2;
         if (flip) flip_image(crop);
-        float exp = rand_uniform(1./exposure, exposure);
-        float sat = rand_uniform(1./saturation, saturation);
-        exposure_image(crop, exp);
-        exposure_image(crop, sat);
+        random_distort_image(crop, hue, saturation, exposure);
 
         /*
         show_image(im, "orig");
@@ -241,6 +238,7 @@ void fill_truth_region(char *path, float *truth, int classes, int num_boxes, int
     labelpath = find_replace(labelpath, "JPEGImages", "labels");
 
     labelpath = find_replace(labelpath, ".jpg", ".txt");
+    labelpath = find_replace(labelpath, ".png", ".txt");
     labelpath = find_replace(labelpath, ".JPG", ".txt");
     labelpath = find_replace(labelpath, ".JPEG", ".txt");
     int count = 0;
@@ -287,6 +285,7 @@ void fill_truth_detection(char *path, int num_boxes, float *truth, int classes,
     labelpath = find_replace(labelpath, "JPEGImages", "labels");
 
     labelpath = find_replace(labelpath, ".jpg", ".txt");
+    labelpath = find_replace(labelpath, ".png", ".txt");
     labelpath = find_replace(labelpath, ".JPG", ".txt");
     labelpath = find_replace(labelpath, ".JPEG", ".txt");
     int count = 0;
@@ -443,7 +442,7 @@ void free_data(data d)
     }
 }
 
-data load_data_region(int n, char **paths, int m, int w, int h, int size, int classes, float jitter)
+data load_data_region(int n, char **paths, int m, int w, int h, int size, int classes, float jitter, float hue, float saturation, float exposure)
 {
     char **random_paths = get_random_paths(paths, n, m);
     int i;
@@ -485,6 +484,7 @@ data load_data_region(int n, char **paths, int m, int w, int h, int size, int cl
 
         image sized = resize_image(cropped, w, h);
         if(flip) flip_image(sized);
+        random_distort_image(sized, hue, saturation, exposure);
         d.X.vals[i] = sized.data;
 
         fill_truth_region(random_paths[i], d.y.vals[i], classes, size, flip, dx, dy, 1./sx, 1./sy);
@@ -611,7 +611,7 @@ data load_data_swag(char **paths, int n, int classes, float jitter)
     return d;
 }
 
-data load_data_detection(int n, char **paths, int m, int w, int h, int boxes, int classes, float jitter)
+data load_data_detection(int n, char **paths, int m, int w, int h, int boxes, int classes, float jitter, float hue, float saturation, float exposure)
 {
     char **random_paths = get_random_paths(paths, n, m);
     int i;
@@ -651,6 +651,7 @@ data load_data_detection(int n, char **paths, int m, int w, int h, int boxes, in
 
         image sized = resize_image(cropped, w, h);
         if(flip) flip_image(sized);
+        random_distort_image(sized, hue, saturation, exposure);
         d.X.vals[i] = sized.data;
 
         fill_truth_detection(random_paths[i], boxes, d.y.vals[i], classes, flip, dx, dy, 1./sx, 1./sy);
@@ -679,17 +680,17 @@ void *load_thread(void *ptr)
     if (a.type == OLD_CLASSIFICATION_DATA){
         *a.d = load_data(a.paths, a.n, a.m, a.labels, a.classes, a.w, a.h);
     } else if (a.type == CLASSIFICATION_DATA){
-        *a.d = load_data_augment(a.paths, a.n, a.m, a.labels, a.classes, a.min, a.max, a.size, a.angle, a.exposure, a.saturation);
+        *a.d = load_data_augment(a.paths, a.n, a.m, a.labels, a.classes, a.min, a.max, a.size, a.angle, a.hue, a.saturation, a.exposure);
     } else if (a.type == SUPER_DATA){
         *a.d = load_data_super(a.paths, a.n, a.m, a.w, a.h, a.scale);
     } else if (a.type == STUDY_DATA){
-        *a.d = load_data_study(a.paths, a.n, a.m, a.labels, a.classes, a.min, a.max, a.size, a.angle, a.exposure, a.saturation);
+        *a.d = load_data_study(a.paths, a.n, a.m, a.labels, a.classes, a.min, a.max, a.size, a.angle, a.hue, a.saturation, a.exposure);
     } else if (a.type == WRITING_DATA){
         *a.d = load_data_writing(a.paths, a.n, a.m, a.w, a.h, a.out_w, a.out_h);
     } else if (a.type == REGION_DATA){
-        *a.d = load_data_region(a.n, a.paths, a.m, a.w, a.h, a.num_boxes, a.classes, a.jitter);
+        *a.d = load_data_region(a.n, a.paths, a.m, a.w, a.h, a.num_boxes, a.classes, a.jitter, a.hue, a.saturation, a.exposure);
     } else if (a.type == DETECTION_DATA){
-        *a.d = load_data_detection(a.n, a.paths, a.m, a.w, a.h, a.num_boxes, a.classes, a.jitter);
+        *a.d = load_data_detection(a.n, a.paths, a.m, a.w, a.h, a.num_boxes, a.classes, a.jitter, a.hue, a.saturation, a.exposure);
     } else if (a.type == SWAG_DATA){
         *a.d = load_data_swag(a.paths, a.n, a.classes, a.jitter);
     } else if (a.type == COMPARE_DATA){
@@ -698,7 +699,7 @@ void *load_thread(void *ptr)
         *(a.im) = load_image_color(a.path, 0, 0);
         *(a.resized) = resize_image(*(a.im), a.w, a.h);
     } else if (a.type == TAG_DATA){
-        *a.d = load_data_tag(a.paths, a.n, a.m, a.classes, a.min, a.max, a.size, a.angle, a.exposure, a.saturation);
+        *a.d = load_data_tag(a.paths, a.n, a.m, a.classes, a.min, a.max, a.size, a.angle, a.hue, a.saturation, a.exposure);
         //*a.d = load_data(a.paths, a.n, a.m, a.labels, a.classes, a.w, a.h);
     }
     free(ptr);
@@ -740,13 +741,13 @@ data load_data(char **paths, int n, int m, char **labels, int k, int w, int h)
     return d;
 }
 
-data load_data_study(char **paths, int n, int m, char **labels, int k, int min, int max, int size, float angle, float exposure, float saturation)
+data load_data_study(char **paths, int n, int m, char **labels, int k, int min, int max, int size, float angle, float hue, float saturation, float exposure)
 {
     data d = {0};
     d.indexes = calloc(n, sizeof(int));
     if(m) paths = get_random_paths_indexes(paths, n, m, d.indexes);
     d.shallow = 0;
-    d.X = load_image_augment_paths(paths, n, min, max, size, angle, exposure, saturation);
+    d.X = load_image_augment_paths(paths, n, min, max, size, angle, hue, saturation, exposure);
     d.y = load_labels_paths(paths, n, labels, k);
     if(m) free(paths);
     return d;
@@ -782,25 +783,25 @@ data load_data_super(char **paths, int n, int m, int w, int h, int scale)
     return d;
 }
 
-data load_data_augment(char **paths, int n, int m, char **labels, int k, int min, int max, int size, float angle, float exposure, float saturation)
+data load_data_augment(char **paths, int n, int m, char **labels, int k, int min, int max, int size, float angle, float hue, float saturation, float exposure)
 {
     if(m) paths = get_random_paths(paths, n, m);
     data d = {0};
     d.shallow = 0;
-    d.X = load_image_augment_paths(paths, n, min, max, size, angle, exposure, saturation);
+    d.X = load_image_augment_paths(paths, n, min, max, size, angle, hue, saturation, exposure);
     d.y = load_labels_paths(paths, n, labels, k);
     if(m) free(paths);
     return d;
 }
 
-data load_data_tag(char **paths, int n, int m, int k, int min, int max, int size, float angle, float exposure, float saturation)
+data load_data_tag(char **paths, int n, int m, int k, int min, int max, int size, float angle, float hue, float saturation, float exposure)
 {
     if(m) paths = get_random_paths(paths, n, m);
     data d = {0};
     d.w = size;
     d.h = size;
     d.shallow = 0;
-    d.X = load_image_augment_paths(paths, n, min, max, size, angle, exposure, saturation);
+    d.X = load_image_augment_paths(paths, n, min, max, size, angle, hue, saturation, exposure);
     d.y = load_tags_paths(paths, n, k);
     if(m) free(paths);
     return d;

src/data.h

@@ -54,6 +54,7 @@ typedef struct load_args{
     float angle;
     float saturation;
     float exposure;
+    float hue;
     data *d;
     image *im;
     image *resized;
@@ -74,11 +75,12 @@ void print_letters(float *pred, int n);
 data load_data_captcha(char **paths, int n, int m, int k, int w, int h);
 data load_data_captcha_encode(char **paths, int n, int m, int w, int h);
 data load_data(char **paths, int n, int m, char **labels, int k, int w, int h);
-data load_data_detection(int n, char **paths, int m, int w, int h, int boxes, int classes, float jitter);
-data load_data_tag(char **paths, int n, int m, int k, int min, int max, int size, float angle, float exposure, float saturation);
-data load_data_augment(char **paths, int n, int m, char **labels, int k, int min, int max, int size, float angle, float exposure, float saturation);
+data load_data_detection(int n, char **paths, int m, int w, int h, int boxes, int classes, float jitter, float hue, float saturation, float exposure);
+data load_data_tag(char **paths, int n, int m, int k, int min, int max, int size, float angle, float hue, float saturation, float exposure);
+matrix load_image_augment_paths(char **paths, int n, int min, int max, int size, float angle, float hue, float saturation, float exposure);
 data load_data_super(char **paths, int n, int m, int w, int h, int scale);
-data load_data_study(char **paths, int n, int m, char **labels, int k, int min, int max, int size, float angle, float exposure, float saturation);
+data load_data_study(char **paths, int n, int m, char **labels, int k, int min, int max, int size, float angle, float hue, float saturation, float exposure);
+data load_data_augment(char **paths, int n, int m, char **labels, int k, int min, int max, int size, float angle, float hue, float saturation, float exposure);
 data load_go(char *filename);
 
 box_label *read_boxes(char *filename, int *n);

src/demo.c

@@ -8,7 +8,7 @@
 #include "demo.h"
 #include <sys/time.h>
 
-#define FRAMES 1
+#define FRAMES 3
 
 #ifdef OPENCV
 #include "opencv2/highgui/highgui_c.h"
@@ -48,7 +48,7 @@ void *fetch_in_thread(void *ptr)
 
 void *detect_in_thread(void *ptr)
 {
-    float nms = .4;
+    float nms = .1;
 
     detection_layer l = net.layers[net.n-1];
     float *X = det_s.data;
@@ -153,6 +153,7 @@ void demo(char *cfgfile, char *weightfile, float thresh, int cam_index, const ch
             if(pthread_create(&fetch_thread, 0, fetch_in_thread, 0)) error("Thread creation failed");
             if(pthread_create(&detect_thread, 0, detect_in_thread, 0)) error("Thread creation failed");
 
+            if(1){
                 show_image(disp, "Demo");
                 int c = cvWaitKey(1);
                 if (c == 10){
@@ -161,6 +162,11 @@ void demo(char *cfgfile, char *weightfile, float thresh, int cam_index, const ch
                     else if(frame_skip == 60) frame_skip = 4;   
                     else frame_skip = 0;
                 }
+            }else{
+                char buff[256];
+                sprintf(buff, "/home/pjreddie/tmp/bag_%07d", count);
+                save_image(disp, buff);
+            }
 
             pthread_join(fetch_thread, 0);
             pthread_join(detect_thread, 0);

src/detection_layer.c

@@ -22,6 +22,8 @@ detection_layer make_detection_layer(int batch, int inputs, int n, int side, int
     l.coords = coords;
     l.rescore = rescore;
     l.side = side;
+    l.w = side;
+    l.h = side;
     assert(side*side*((1 + l.coords)*l.n + l.classes) == inputs);
     l.cost = calloc(1, sizeof(float));
     l.outputs = l.inputs;
@@ -44,6 +46,7 @@ void forward_detection_layer(const detection_layer l, network_state state)
     int locations = l.side*l.side;
     int i,j;
     memcpy(l.output, state.input, l.outputs*l.batch*sizeof(float));
+    //if(l.reorg) reorg(l.output, l.w*l.h, size*l.n, l.batch, 1);
     int b;
     if (l.softmax){
         for(b = 0; b < l.batch; ++b){
@@ -204,6 +207,7 @@ void forward_detection_layer(const detection_layer l, network_state state)
 
 
         printf("Detection Avg IOU: %f, Pos Cat: %f, All Cat: %f, Pos Obj: %f, Any Obj: %f, count: %d\n", avg_iou/count, avg_cat/count, avg_allcat/(count*l.classes), avg_obj/count, avg_anyobj/(l.batch*locations*l.n), count);
+        //if(l.reorg) reorg(l.delta, l.w*l.h, size*l.n, l.batch, 0);
     }
 }
 

src/detector.c

@@ -51,6 +51,11 @@ void train_detector(char *cfgfile, char *weightfile)
     args.d = &buffer;
     args.type = DETECTION_DATA;
 
+    args.angle = net.angle;
+    args.exposure = net.exposure;
+    args.saturation = net.saturation;
+    args.hue = net.hue;
+
     pthread_t load_thread = load_data_in_thread(args);
     clock_t time;
     //while(i*imgs < N*120){

src/im2col.c

@@ -18,13 +18,9 @@ void im2col_cpu(float* data_im,
      int ksize,  int stride, int pad, float* data_col) 
 {
     int c,h,w;
-    int height_col = (height - ksize) / stride + 1;
-    int width_col = (width - ksize) / stride + 1;
-    if (pad){
-        height_col = 1 + (height-1) / stride;
-        width_col = 1 + (width-1) / stride;
-        pad = ksize/2;
-    }
+    int height_col = (height + 2*pad - ksize) / stride + 1;
+    int width_col = (width + 2*pad - ksize) / stride + 1;
+
     int channels_col = channels * ksize * ksize;
     for (c = 0; c < channels_col; ++c) {
         int w_offset = c % ksize;

src/im2col_kernels.cu

@@ -33,8 +33,12 @@ __global__ void im2col_gpu_kernel(const int n, const float* data_im,
             for (int j = 0; j < ksize; ++j) {
                 int h = h_in + i;
                 int w = w_in + j;
+
                 *data_col_ptr = (h >= 0 && w >= 0 && h < height && w < width) ?
                     data_im_ptr[i * width + j] : 0;
+
+                //*data_col_ptr = data_im_ptr[ii * width + jj];
+
                 data_col_ptr += height_col * width_col;
             }
         }
@@ -46,7 +50,6 @@ void im2col_ongpu(float *im,
          int ksize, int stride, int pad, float *data_col){
     // We are going to launch channels * height_col * width_col kernels, each
     // kernel responsible for copying a single-channel grid.
-    pad = pad ? ksize/2 : 0;
     int height_col = (height + 2 * pad - ksize) / stride + 1;
     int width_col = (width + 2 * pad - ksize) / stride + 1;
     int num_kernels = channels * height_col * width_col;
@@ -56,93 +59,3 @@ void im2col_ongpu(float *im,
                 stride, height_col,
                 width_col, data_col);
 }
-/*
-   __global__ void im2col_pad_kernel(float *im,
-   int channels,  int height,  int width,
-   int ksize,  int stride, float *data_col)
-   {
-   int c,h,w;
-   int height_col = 1 + (height-1) / stride;
-   int width_col = 1 + (width-1) / stride;
-   int channels_col = channels * ksize * ksize;
-
-   int pad = ksize/2;
-
-   int id = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
-   int col_size = height_col*width_col*channels_col;
-   if (id >= col_size) return;
-
-   int col_index = id;
-   w = id % width_col;
-   id /= width_col;
-   h = id % height_col;
-   id /= height_col;
-   c = id % channels_col;
-   id /= channels_col;
-
-   int w_offset = c % ksize;
-   int h_offset = (c / ksize) % ksize;
-   int im_channel = c / ksize / ksize;
-   int im_row = h_offset + h * stride - pad;
-   int im_col = w_offset + w * stride - pad;
-
-   int im_index = im_col + width*(im_row + height*im_channel);
-   float val = (im_row < 0 || im_col < 0 || im_row >= height || im_col >= width) ? 0 : im[im_index];
-
-   data_col[col_index] = val;
-   }
-
-   __global__ void im2col_nopad_kernel(float *im,
-   int channels,  int height,  int width,
-   int ksize,  int stride, float *data_col)
-   {
-   int c,h,w;
-   int height_col = (height - ksize) / stride + 1;
-   int width_col = (width - ksize) / stride + 1;
-   int channels_col = channels * ksize * ksize;
-
-   int id = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
-   int col_size = height_col*width_col*channels_col;
-   if (id >= col_size) return;
-
-   int col_index = id;
-   w = id % width_col;
-   id /= width_col;
-   h = id % height_col;
-   id /= height_col;
-   c = id % channels_col;
-   id /= channels_col;
-
-   int w_offset = c % ksize;
-   int h_offset = (c / ksize) % ksize;
-   int im_channel = c / ksize / ksize;
-   int im_row = h_offset + h * stride;
-   int im_col = w_offset + w * stride;
-
-   int im_index = im_col + width*(im_row + height*im_channel);
-   float val = (im_row < 0 || im_col < 0 || im_row >= height || im_col >= width) ? 0 : im[im_index];
-
-   data_col[col_index] = val;
-   }
-
-   extern "C" void im2col_ongpu(float *im,
-   int channels,  int height,  int width,
-int ksize,  int stride,  int pad, float *data_col)
-{
-
-    int height_col = (height - ksize) / stride + 1;
-    int width_col = (width - ksize) / stride + 1;
-    int channels_col = channels * ksize * ksize;
-
-    if (pad){
-        height_col = 1 + (height-1) / stride;
-        width_col = 1 + (width-1) / stride;
-    }
-
-    size_t n = channels_col*height_col*width_col;
-
-    if(pad)im2col_pad_kernel<<<cuda_gridsize(n),BLOCK>>>(im,  channels, height, width, ksize, stride, data_col);
-    else im2col_nopad_kernel<<<cuda_gridsize(n),BLOCK>>>(im,  channels, height, width, ksize, stride, data_col);
-    check_error(cudaPeekAtLastError());
-}
-*/

src/image.c

@@ -1,6 +1,7 @@
 #include "image.h"
 #include "utils.h"
 #include "blas.h"
+#include "cuda.h"
 #include <stdio.h>
 #include <math.h>
 
@@ -247,6 +248,26 @@ void constrain_image(image im)
 }
 
 void normalize_image(image p)
+{
+    int i;
+    float min = 9999999;
+    float max = -999999;
+
+    for(i = 0; i < p.h*p.w*p.c; ++i){
+        float v = p.data[i];
+        if(v < min) min = v;
+        if(v > max) max = v;
+    }
+    if(max - min < .000000001){
+        min = 0;
+        max = 1;
+    }
+    for(i = 0; i < p.c*p.w*p.h; ++i){
+        p.data[i] = (p.data[i] - min)/(max-min);
+    }
+}
+
+void normalize_image2(image p)
 {
     float *min = calloc(p.c, sizeof(float));
     float *max = calloc(p.c, sizeof(float));
@@ -320,7 +341,6 @@ void show_image_cv(image p, const char *name)
     }
     free_image(copy);
     if(0){
-        //if(disp->height < 448 || disp->width < 448 || disp->height > 1000){
         int w = 448;
         int h = w*p.h/p.w;
         if(h > 1000){
@@ -334,35 +354,35 @@ void show_image_cv(image p, const char *name)
     }
     cvShowImage(buff, disp);
     cvReleaseImage(&disp);
-    }
+}
 #endif
 
-    void show_image(image p, const char *name)
-    {
+void show_image(image p, const char *name)
+{
 #ifdef OPENCV
     show_image_cv(p, name);
 #else
     fprintf(stderr, "Not compiled with OpenCV, saving to %s.png instead\n", name);
     save_image(p, name);
 #endif
-    }
+}
 
 #ifdef OPENCV
-    image get_image_from_stream(CvCapture *cap)
-    {
+image get_image_from_stream(CvCapture *cap)
+{
     IplImage* src = cvQueryFrame(cap);
     if (!src) return make_empty_image(0,0,0);
     image im = ipl_to_image(src);
     rgbgr_image(im);
     return im;
-    }
+}
 #endif
 
 #ifdef OPENCV
-    void save_image_jpg(image p, const char *name)
-    {
+void save_image_jpg(image p, const char *name)
+{
     image copy = copy_image(p);
-        rgbgr_image(copy);
+    if(p.c == 3) rgbgr_image(copy);
     int x,y,k;
 
     char buff[256];
@@ -380,14 +400,14 @@ void show_image_cv(image p, const char *name)
     cvSaveImage(buff, disp,0);
     cvReleaseImage(&disp);
     free_image(copy);
-    }
+}
 #endif
 
-    void save_image(image im, const char *name)
-    {
-        #ifdef OPENCV
+void save_image(image im, const char *name)
+{
+#ifdef OPENCV
     save_image_jpg(im, name);
-        #else
+#else
     char buff[256];
     //sprintf(buff, "%s (%d)", name, windows);
     sprintf(buff, "%s.png", name);
@@ -401,12 +421,12 @@ void show_image_cv(image p, const char *name)
     int success = stbi_write_png(buff, im.w, im.h, im.c, data, im.w*im.c);
     free(data);
     if(!success) fprintf(stderr, "Failed to write image %s\n", buff);
-        #endif
-    }
+#endif
+}
 
 
-    void show_image_layers(image p, char *name)
-    {
+void show_image_layers(image p, char *name)
+{
     int i;
     char buff[256];
     for(i = 0; i < p.c; ++i){
@@ -415,34 +435,34 @@ void show_image_cv(image p, const char *name)
         show_image(layer, buff);
         free_image(layer);
     }
-    }
+}
 
-    void show_image_collapsed(image p, char *name)
-    {
+void show_image_collapsed(image p, char *name)
+{
     image c = collapse_image_layers(p, 1);
     show_image(c, name);
     free_image(c);
-    }
+}
 
-    image make_empty_image(int w, int h, int c)
-    {
+image make_empty_image(int w, int h, int c)
+{
     image out;
     out.data = 0;
     out.h = h;
     out.w = w;
     out.c = c;
     return out;
-    }
+}
 
-    image make_image(int w, int h, int c)
-    {
+image make_image(int w, int h, int c)
+{
     image out = make_empty_image(w,h,c);
     out.data = calloc(h*w*c, sizeof(float));
     return out;
-    }
+}
 
-    image make_random_image(int w, int h, int c)
-    {
+image make_random_image(int w, int h, int c)
+{
     image out = make_empty_image(w,h,c);
     out.data = calloc(h*w*c, sizeof(float));
     int i;
@@ -450,17 +470,17 @@ void show_image_cv(image p, const char *name)
         out.data[i] = (rand_normal() * .25) + .5;
     }
     return out;
-    }
+}
 
-    image float_to_image(int w, int h, int c, float *data)
-    {
+image float_to_image(int w, int h, int c, float *data)
+{
     image out = make_empty_image(w,h,c);
     out.data = data;
     return out;
-    }
+}
 
-    image rotate_crop_image(image im, float rad, float s, int w, int h, int dx, int dy)
-    {
+image rotate_crop_image(image im, float rad, float s, int w, int h, int dx, int dy)
+{
     int x, y, c;
     float cx = im.w/2.;
     float cy = im.h/2.;
@@ -476,10 +496,10 @@ void show_image_cv(image p, const char *name)
         }
     }
     return rot;
-    }
+}
 
-    image rotate_image(image im, float rad)
-    {
+image rotate_image(image im, float rad)
+{
     int x, y, c;
     float cx = im.w/2.;
     float cy = im.h/2.;
@@ -495,22 +515,22 @@ void show_image_cv(image p, const char *name)
         }
     }
     return rot;
-    }
+}
 
-    void translate_image(image m, float s)
-    {
+void translate_image(image m, float s)
+{
     int i;
     for(i = 0; i < m.h*m.w*m.c; ++i) m.data[i] += s;
-    }
+}
 
-    void scale_image(image m, float s)
-    {
+void scale_image(image m, float s)
+{
     int i;
     for(i = 0; i < m.h*m.w*m.c; ++i) m.data[i] *= s;
-    }
+}
 
-    image crop_image(image im, int dx, int dy, int w, int h)
-    {
+image crop_image(image im, int dx, int dy, int w, int h)
+{
     image cropped = make_image(w, h, im.c);
     int i, j, k;
     for(k = 0; k < im.c; ++k){
@@ -529,7 +549,7 @@ void show_image_cv(image p, const char *name)
         }
     }
     return cropped;
-    }
+}
 
 int best_3d_shift_r(image a, image b, int min, int max)
 {
@@ -666,7 +686,7 @@ void rgb_to_hsv(image im)
             v = max;
             if(max == 0){
                 s = 0;
-                h = -1;
+                h = 0;
             }else{
                 s = delta/max;
                 if(r == max){
@@ -677,6 +697,7 @@ void rgb_to_hsv(image im)
                     h = 4 + (r - g) / delta;
                 }
                 if (h < 0) h += 6;
+                h = h/6.;
             }
             set_pixel(im, i, j, 0, h);
             set_pixel(im, i, j, 1, s);
@@ -694,7 +715,7 @@ void hsv_to_rgb(image im)
     float f, p, q, t;
     for(j = 0; j < im.h; ++j){
         for(i = 0; i < im.w; ++i){
-            h = get_pixel(im, i , j, 0);
+            h = 6 * get_pixel(im, i , j, 0);
             s = get_pixel(im, i , j, 1);
             v = get_pixel(im, i , j, 2);
             if (s == 0) {
@@ -781,6 +802,18 @@ void scale_image_channel(image im, int c, float v)
     }
 }
 
+void translate_image_channel(image im, int c, float v)
+{
+    int i, j;
+    for(j = 0; j < im.h; ++j){
+        for(i = 0; i < im.w; ++i){
+            float pix = get_pixel(im, i, j, c);
+            pix = pix+v;
+            set_pixel(im, i, j, c, pix);
+        }
+    }
+}
+
 image binarize_image(image im)
 {
     image c = copy_image(im);
@@ -800,6 +833,19 @@ void saturate_image(image im, float sat)
     constrain_image(im);
 }
 
+void hue_image(image im, float hue)
+{
+    rgb_to_hsv(im);
+    int i;
+    for(i = 0; i < im.w*im.h; ++i){
+        im.data[i] = im.data[i] + hue;
+        if (im.data[i] > 1) im.data[i] -= 1;
+        if (im.data[i] < 0) im.data[i] += 1;
+    }
+    hsv_to_rgb(im);
+    constrain_image(im);
+}
+
 void exposure_image(image im, float sat)
 {
     rgb_to_hsv(im);
@@ -808,6 +854,29 @@ void exposure_image(image im, float sat)
     constrain_image(im);
 }
 
+void distort_image(image im, float hue, float sat, float val)
+{
+    rgb_to_hsv(im);
+    scale_image_channel(im, 1, sat);
+    scale_image_channel(im, 2, val);
+    int i;
+    for(i = 0; i < im.w*im.h; ++i){
+        im.data[i] = im.data[i] + hue;
+        if (im.data[i] > 1) im.data[i] -= 1;
+        if (im.data[i] < 0) im.data[i] += 1;
+    }
+    hsv_to_rgb(im);
+    constrain_image(im);
+}
+
+void random_distort_image(image im, float hue, float saturation, float exposure)
+{
+    float dhue = rand_uniform(-hue, hue);
+    float dsat = rand_scale(saturation);
+    float dexp = rand_scale(exposure);
+    distort_image(im, dhue, dsat, dexp);
+}
+
 void saturate_exposure_image(image im, float sat, float exposure)
 {
     rgb_to_hsv(im);
@@ -876,7 +945,6 @@ image resize_image(image im, int w, int h)
     return resized;
 }
 
-#include "cuda.h"
 
 void test_resize(char *filename)
 {
@@ -885,59 +953,40 @@ void test_resize(char *filename)
     printf("L2 Norm: %f\n", mag);
     image gray = grayscale_image(im);
 
-    image sat2 = copy_image(im);
-    saturate_image(sat2, 2);
+    image c1 = copy_image(im);
+    image c2 = copy_image(im);
+    image c3 = copy_image(im);
+    image c4 = copy_image(im);
+    distort_image(c1, .1, 1.5, 1.5);
+    distort_image(c2, -.1, .66666, .66666);
+    distort_image(c3, .1, 1.5, .66666);
+    distort_image(c4, .1, .66666, 1.5);
 
-    image sat5 = copy_image(im);
-    saturate_image(sat5, .5);
 
-    image exp2 = copy_image(im);
-    exposure_image(exp2, 2);
-
-    image exp5 = copy_image(im);
-    exposure_image(exp5, .5);
-
-    image bin = binarize_image(im);
-
-/*
-#ifdef GPU
-    image r = resize_image(im, im.w, im.h);
-    image black = make_image(im.w*2 + 3, im.h*2 + 3, 9);
-    image black2 = make_image(im.w, im.h, 3);
-
-    float *r_gpu = cuda_make_array(r.data, r.w*r.h*r.c);
-    float *black_gpu = cuda_make_array(black.data, black.w*black.h*black.c);
-    float *black2_gpu = cuda_make_array(black2.data, black2.w*black2.h*black2.c);
-    shortcut_gpu(3, r.w, r.h, 1, r_gpu, black.w, black.h, 3, black_gpu);
-    //flip_image(r);
-    //shortcut_gpu(3, r.w, r.h, 1, r.data, black.w, black.h, 3, black.data);
-
-    shortcut_gpu(3, black.w, black.h, 3, black_gpu, black2.w, black2.h, 1, black2_gpu);
-    cuda_pull_array(black_gpu, black.data, black.w*black.h*black.c);
-    cuda_pull_array(black2_gpu, black2.data, black2.w*black2.h*black2.c);
-    show_image_layers(black, "Black");
-    show_image(black2, "Recreate");
-#endif
-*/
-    image rot = rotate_crop_image(im, -.2618, 1, im.w/2, im.h/2, 0, 0);
-    image rot3 = rotate_crop_image(im, -.2618, 2, im.w, im.h, im.w/2, 0);
-    image rot2 = rotate_crop_image(im, -.2618, 1, im.w, im.h, 0, 0);
-    show_image(rot, "Rotated");
-    show_image(rot2, "base");
-
-    show_image(rot3, "Rotated2");
-
-/*
     show_image(im,   "Original");
-    show_image(bin,  "Binary");
     show_image(gray, "Gray");
-    show_image(sat2, "Saturation-2");
-    show_image(sat5, "Saturation-.5");
-    show_image(exp2, "Exposure-2");
-    show_image(exp5, "Exposure-.5");
-    */
+    show_image(c1, "C1");
+    show_image(c2, "C2");
+    show_image(c3, "C3");
+    show_image(c4, "C4");
 #ifdef OPENCV
+    while(1){
+        float exposure = 1.15;
+        float saturation = 1.15;
+        float hue = .05;
+
+        image c = copy_image(im);
+
+        float dexp = rand_scale(exposure);
+        float dsat = rand_scale(saturation);
+        float dhue = rand_uniform(-hue, hue);
+
+        distort_image(c, dhue, dsat, dexp);
+        show_image(c, "rand");
+        printf("%f %f %f\n", dhue, dsat, dexp);
+        free_image(c);
         cvWaitKey(0);
+    }
 #endif
 }
 
@@ -1180,10 +1229,8 @@ void show_images(image *ims, int n, char *window)
        image sized = resize_image(m, w, h);
      */
     normalize_image(m);
-    image sized = resize_image(m, m.w, m.h);
-    save_image(sized, window);
-    show_image(sized, window);
-    free_image(sized);
+    save_image(m, window);
+    show_image(m, window);
     free_image(m);
 }
 

src/image.h

@@ -32,6 +32,7 @@ void scale_image(image m, float s);
 image crop_image(image im, int dx, int dy, int w, int h);
 image random_crop_image(image im, int w, int h);
 image random_augment_image(image im, float angle, int low, int high, int size);
+void random_distort_image(image im, float hue, float saturation, float exposure);
 image resize_image(image im, int w, int h);
 image resize_min(image im, int min);
 void translate_image(image m, float s);
@@ -41,6 +42,7 @@ void rotate_image_cw(image im, int times);
 void embed_image(image source, image dest, int dx, int dy);
 void saturate_image(image im, float sat);
 void exposure_image(image im, float sat);
+void distort_image(image im, float hue, float sat, float val);
 void saturate_exposure_image(image im, float sat, float exposure);
 void hsv_to_rgb(image im);
 void rgbgr_image(image im);

src/imagenet.c

@@ -1,237 +0,0 @@
-#include "network.h"
-#include "utils.h"
-#include "parser.h"
-
-#ifdef OPENCV
-#include "opencv2/highgui/highgui_c.h"
-#endif
-
-void train_imagenet(char *cfgfile, char *weightfile)
-{
-    data_seed = time(0);
-    srand(time(0));
-    float avg_loss = -1;
-    char *base = basecfg(cfgfile);
-    char *backup_directory = "/home/pjreddie/backup/";
-    printf("%s\n", base);
-    network net = parse_network_cfg(cfgfile);
-    if(weightfile){
-        load_weights(&net, weightfile);
-    }
-    printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
-    int imgs = 1024;
-    char **labels = get_labels("data/inet.labels.list");
-    list *plist = get_paths("data/inet.train.list");
-    char **paths = (char **)list_to_array(plist);
-    printf("%d\n", plist->size);
-    int N = plist->size;
-    clock_t time;
-    pthread_t load_thread;
-    data train;
-    data buffer;
-
-    load_args args = {0};
-    args.w = net.w;
-    args.h = net.h;
-    args.paths = paths;
-    args.classes = 1000;
-    args.n = imgs;
-    args.m = N;
-    args.labels = labels;
-    args.d = &buffer;
-    args.type = OLD_CLASSIFICATION_DATA;
-
-    load_thread = load_data_in_thread(args);
-    int epoch = (*net.seen)/N;
-    while(get_current_batch(net) < net.max_batches || net.max_batches == 0){
-        time=clock();
-        pthread_join(load_thread, 0);
-        train = buffer;
-
-        load_thread = load_data_in_thread(args);
-        printf("Loaded: %lf seconds\n", sec(clock()-time));
-        time=clock();
-        float loss = train_network(net, train);
-        if(avg_loss == -1) avg_loss = loss;
-        avg_loss = avg_loss*.9 + loss*.1;
-        printf("%d, %.3f: %f, %f avg, %f rate, %lf seconds, %d images\n", get_current_batch(net), (float)(*net.seen)/N, loss, avg_loss, get_current_rate(net), sec(clock()-time), *net.seen);
-        free_data(train);
-        if(*net.seen/N > epoch){
-            epoch = *net.seen/N;
-            char buff[256];
-            sprintf(buff, "%s/%s_%d.weights",backup_directory,base, epoch);
-            save_weights(net, buff);
-        }
-        if(*net.seen%1000 == 0){
-            char buff[256];
-            sprintf(buff, "%s/%s.backup",backup_directory,base);
-            save_weights(net, buff);
-        }
-    }
-    char buff[256];
-    sprintf(buff, "%s/%s.weights", backup_directory, base);
-    save_weights(net, buff);
-
-    pthread_join(load_thread, 0);
-    free_data(buffer);
-    free_network(net);
-    free_ptrs((void**)labels, 1000);
-    free_ptrs((void**)paths, plist->size);
-    free_list(plist);
-    free(base);
-}
-
-void validate_imagenet(char *filename, char *weightfile)
-{
-    int i = 0;
-    network net = parse_network_cfg(filename);
-    if(weightfile){
-        load_weights(&net, weightfile);
-    }
-    srand(time(0));
-
-    char **labels = get_labels("data/inet.labels.list");
-    //list *plist = get_paths("data/inet.suppress.list");
-    list *plist = get_paths("data/inet.val.list");
-
-    char **paths = (char **)list_to_array(plist);
-    int m = plist->size;
-    free_list(plist);
-
-    clock_t time;
-    float avg_acc = 0;
-    float avg_top5 = 0;
-    int splits = 50;
-    int num = (i+1)*m/splits - i*m/splits;
-
-    data val, buffer;
-
-    load_args args = {0};
-    args.w = net.w;
-    args.h = net.h;
-    args.paths = paths;
-    args.classes = 1000;
-    args.n = num;
-    args.m = 0;
-    args.labels = labels;
-    args.d = &buffer;
-    args.type = OLD_CLASSIFICATION_DATA;
-
-    pthread_t load_thread = load_data_in_thread(args);
-    for(i = 1; i <= splits; ++i){
-        time=clock();
-
-        pthread_join(load_thread, 0);
-        val = buffer;
-
-        num = (i+1)*m/splits - i*m/splits;
-        char **part = paths+(i*m/splits);
-        if(i != splits){
-            args.paths = part;
-            load_thread = load_data_in_thread(args);
-        }
-        printf("Loaded: %d images in %lf seconds\n", val.X.rows, sec(clock()-time));
-
-        time=clock();
-        float *acc = network_accuracies(net, val, 5);
-        avg_acc += acc[0];
-        avg_top5 += acc[1];
-        printf("%d: top1: %f, top5: %f, %lf seconds, %d images\n", i, avg_acc/i, avg_top5/i, sec(clock()-time), val.X.rows);
-        free_data(val);
-    }
-}
-
-void test_imagenet(char *cfgfile, char *weightfile, char *filename)
-{
-    network net = parse_network_cfg(cfgfile);
-    if(weightfile){
-        load_weights(&net, weightfile);
-    }
-    set_batch_network(&net, 1);
-    srand(2222222);
-    int i = 0;
-    char **names = get_labels("data/shortnames.txt");
-    clock_t time;
-    int indexes[10];
-    char buff[256];
-    char *input = buff;
-    while(1){
-        if(filename){
-            strncpy(input, filename, 256);
-        }else{
-            printf("Enter Image Path: ");
-            fflush(stdout);
-            input = fgets(input, 256, stdin);
-            if(!input) return;
-            strtok(input, "\n");
-        }
-        image im = load_image_color(input, 256, 256);
-        float *X = im.data;
-        time=clock();
-        float *predictions = network_predict(net, X);
-        top_predictions(net, 10, indexes);
-        printf("%s: Predicted in %f seconds.\n", input, sec(clock()-time));
-        for(i = 0; i < 10; ++i){
-            int index = indexes[i];
-            printf("%s: %f\n", names[index], predictions[index]);
-        }
-        free_image(im);
-        if (filename) break;
-    }
-}
-
-void run_imagenet(int argc, char **argv)
-{
-    if(argc < 4){
-        fprintf(stderr, "usage: %s %s [train/test/valid] [cfg] [weights (optional)]\n", argv[0], argv[1]);
-        return;
-    }
-
-    char *cfg = argv[3];
-    char *weights = (argc > 4) ? argv[4] : 0;
-    char *filename = (argc > 5) ? argv[5]: 0;
-    if(0==strcmp(argv[2], "test")) test_imagenet(cfg, weights, filename);
-    else if(0==strcmp(argv[2], "train")) train_imagenet(cfg, weights);
-    else if(0==strcmp(argv[2], "valid")) validate_imagenet(cfg, weights);
-}
-
-/*
-   void train_imagenet_distributed(char *address)
-   {
-   float avg_loss = 1;
-   srand(time(0));
-   network net = parse_network_cfg("cfg/net.cfg");
-   set_learning_network(&net, 0, 1, 0);
-   printf("Learning Rate: %g, Momentum: %g, Decay: %g\n", net.learning_rate, net.momentum, net.decay);
-   int imgs = net.batch;
-   int i = 0;
-   char **labels = get_labels("/home/pjreddie/data/imagenet/cls.labels.list");
-   list *plist = get_paths("/data/imagenet/cls.train.list");
-   char **paths = (char **)list_to_array(plist);
-   printf("%d\n", plist->size);
-   clock_t time;
-   data train, buffer;
-   pthread_t load_thread = load_data_thread(paths, imgs, plist->size, labels, 1000, 224, 224, &buffer);
-   while(1){
-   i += 1;
-
-   time=clock();
-   client_update(net, address);
-   printf("Updated: %lf seconds\n", sec(clock()-time));
-
-   time=clock();
-   pthread_join(load_thread, 0);
-   train = buffer;
-   normalize_data_rows(train);
-   load_thread = load_data_thread(paths, imgs, plist->size, labels, 1000, 224, 224, &buffer);
-   printf("Loaded: %lf seconds\n", sec(clock()-time));
-   time=clock();
-
-   float loss = train_network(net, train);
-   avg_loss = avg_loss*.9 + loss*.1;
-   printf("%d: %f, %f avg, %lf seconds, %d images\n", i, loss, avg_loss, sec(clock()-time), i*imgs);
-   free_data(train);
-   }
-   }
- */
-

src/layer.h

@@ -72,6 +72,7 @@ struct layer{
     float saturation;
     float exposure;
     float shift;
+    float ratio;
     int softmax;
     int classes;
     int coords;
@@ -82,6 +83,7 @@ struct layer{
     int joint;
     int noadjust;
     int reorg;
+    int log;
 
     float alpha;
     float beta;

src/maxpool_layer.c

@@ -18,7 +18,7 @@ image get_maxpool_delta(maxpool_layer l)
     return float_to_image(w,h,c,l.delta);
 }
 
-maxpool_layer make_maxpool_layer(int batch, int h, int w, int c, int size, int stride)
+maxpool_layer make_maxpool_layer(int batch, int h, int w, int c, int size, int stride, int padding)
 {
     fprintf(stderr, "Maxpool Layer: %d x %d x %d image, %d size, %d stride\n", h,w,c,size,stride);
     maxpool_layer l = {0};
@@ -27,8 +27,9 @@ maxpool_layer make_maxpool_layer(int batch, int h, int w, int c, int size, int s
     l.h = h;
     l.w = w;
     l.c = c;
-    l.out_w = (w-1)/stride + 1;
-    l.out_h = (h-1)/stride + 1;
+    l.pad = padding;
+    l.out_w = (w + 2*padding - size + 1)/stride + 1;
+    l.out_h = (h + 2*padding - size + 1)/stride + 1;
     l.out_c = c;
     l.outputs = l.out_h * l.out_w * l.out_c;
     l.inputs = h*w*c;
@@ -48,13 +49,12 @@ maxpool_layer make_maxpool_layer(int batch, int h, int w, int c, int size, int s
 
 void resize_maxpool_layer(maxpool_layer *l, int w, int h)
 {
-    int stride = l->stride;
     l->h = h;
     l->w = w;
     l->inputs = h*w*l->c;
 
-    l->out_w = (w-1)/stride + 1;
-    l->out_h = (h-1)/stride + 1;
+    l->out_w = (w + 2*l->pad - l->size + 1)/l->stride + 1;
+    l->out_h = (h + 2*l->pad - l->size + 1)/l->stride + 1;
     l->outputs = l->out_w * l->out_h * l->c;
     int output_size = l->outputs * l->batch;
 
@@ -75,11 +75,11 @@ void resize_maxpool_layer(maxpool_layer *l, int w, int h)
 void forward_maxpool_layer(const maxpool_layer l, network_state state)
 {
     int b,i,j,k,m,n;
-    int w_offset = (-l.size-1)/2 + 1;
-    int h_offset = (-l.size-1)/2 + 1;
+    int w_offset = -l.pad;
+    int h_offset = -l.pad;
 
-    int h = (l.h-1)/l.stride + 1;
-    int w = (l.w-1)/l.stride + 1;
+    int h = l.out_h;
+    int w = l.out_w;
     int c = l.c;
 
     for(b = 0; b < l.batch; ++b){
@@ -112,8 +112,8 @@ void forward_maxpool_layer(const maxpool_layer l, network_state state)
 void backward_maxpool_layer(const maxpool_layer l, network_state state)
 {
     int i;
-    int h = (l.h-1)/l.stride + 1;
-    int w = (l.w-1)/l.stride + 1;
+    int h = l.out_h;
+    int w = l.out_w;
     int c = l.c;
     for(i = 0; i < h*w*c*l.batch; ++i){
         int index = l.indexes[i];

src/maxpool_layer.h

@@ -9,7 +9,7 @@
 typedef layer maxpool_layer;
 
 image get_maxpool_image(maxpool_layer l);
-maxpool_layer make_maxpool_layer(int batch, int h, int w, int c, int size, int stride);
+maxpool_layer make_maxpool_layer(int batch, int h, int w, int c, int size, int stride, int padding);
 void resize_maxpool_layer(maxpool_layer *l, int w, int h);
 void forward_maxpool_layer(const maxpool_layer l, network_state state);
 void backward_maxpool_layer(const maxpool_layer l, network_state state);

src/maxpool_layer_kernels.cu

@@ -7,10 +7,10 @@ extern "C" {
 #include "cuda.h"
 }
 
-__global__ void forward_maxpool_layer_kernel(int n, int in_h, int in_w, int in_c, int stride, int size, float *input, float *output, int *indexes)
+__global__ void forward_maxpool_layer_kernel(int n, int in_h, int in_w, int in_c, int stride, int size, int pad, float *input, float *output, int *indexes)
 {
-    int h = (in_h-1)/stride + 1;
-    int w = (in_w-1)/stride + 1;
+    int h = (in_h + 2*pad - size + 1)/stride + 1;
+    int w = (in_w + 2*pad - size + 1)/stride + 1;
     int c = in_c;
 
     int id = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
@@ -24,8 +24,8 @@ __global__ void forward_maxpool_layer_kernel(int n, int in_h, int in_w, int in_c
     id /= c;
     int b = id;
 
-    int w_offset = (-size-1)/2 + 1;
-    int h_offset = (-size-1)/2 + 1;
+    int w_offset = -pad;
+    int h_offset = -pad;
 
     int out_index = j + w*(i + h*(k + c*b));
     float max = -INFINITY;
@@ -47,10 +47,10 @@ __global__ void forward_maxpool_layer_kernel(int n, int in_h, int in_w, int in_c
     indexes[out_index] = max_i;
 }
 
-__global__ void backward_maxpool_layer_kernel(int n, int in_h, int in_w, int in_c, int stride, int size, float *delta, float *prev_delta, int *indexes)
+__global__ void backward_maxpool_layer_kernel(int n, int in_h, int in_w, int in_c, int stride, int size, int pad, float *delta, float *prev_delta, int *indexes)
 {
-    int h = (in_h-1)/stride + 1;
-    int w = (in_w-1)/stride + 1;
+    int h = (in_h + 2*pad - size + 1)/stride + 1;
+    int w = (in_w + 2*pad - size + 1)/stride + 1;
     int c = in_c;
     int area = (size-1)/stride;
 
@@ -66,8 +66,8 @@ __global__ void backward_maxpool_layer_kernel(int n, int in_h, int in_w, int in_
     id /= in_c;
     int b = id;
 
-    int w_offset = (-size-1)/2 + 1;
-    int h_offset = (-size-1)/2 + 1;
+    int w_offset = -pad;
+    int h_offset = -pad;
 
     float d = 0;
     int l, m;
@@ -86,13 +86,13 @@ __global__ void backward_maxpool_layer_kernel(int n, int in_h, int in_w, int in_
 
 extern "C" void forward_maxpool_layer_gpu(maxpool_layer layer, network_state state)
 {
-    int h = (layer.h-1)/layer.stride + 1;
-    int w = (layer.w-1)/layer.stride + 1;
+    int h = layer.out_h;
+    int w = layer.out_w;
     int c = layer.c;
 
     size_t n = h*w*c*layer.batch;
 
-    forward_maxpool_layer_kernel<<<cuda_gridsize(n), BLOCK>>>(n, layer.h, layer.w, layer.c, layer.stride, layer.size, state.input, layer.output_gpu, layer.indexes_gpu);
+    forward_maxpool_layer_kernel<<<cuda_gridsize(n), BLOCK>>>(n, layer.h, layer.w, layer.c, layer.stride, layer.size, layer.pad, state.input, layer.output_gpu, layer.indexes_gpu);
     check_error(cudaPeekAtLastError());
 }
 
@@ -100,7 +100,7 @@ extern "C" void backward_maxpool_layer_gpu(maxpool_layer layer, network_state st
 {
     size_t n = layer.h*layer.w*layer.c*layer.batch;
 
-    backward_maxpool_layer_kernel<<<cuda_gridsize(n), BLOCK>>>(n, layer.h, layer.w, layer.c, layer.stride, layer.size, layer.delta_gpu, state.delta, layer.indexes_gpu);
+    backward_maxpool_layer_kernel<<<cuda_gridsize(n), BLOCK>>>(n, layer.h, layer.w, layer.c, layer.stride, layer.size, layer.pad, layer.delta_gpu, state.delta, layer.indexes_gpu);
     check_error(cudaPeekAtLastError());
 }
 

src/network.c

@@ -420,7 +420,7 @@ int resize_network(network *net, int w, int h)
     net->h = h;
     int inputs = 0;
     size_t workspace_size = 0;
-    //fprintf(stderr, "Resizing to %d x %d...", w, h);
+    //fprintf(stderr, "Resizing to %d x %d...\n", w, h);
     //fflush(stderr);
     for (i = 0; i < net->n; ++i){
         layer l = net->layers[i];

src/network.h

@@ -43,6 +43,7 @@ typedef struct network{
     float angle;
     float exposure;
     float saturation;
+    float hue;
 
     int gpu_index;
 

src/parser.c

@@ -2,6 +2,7 @@
 #include <string.h>
 #include <stdlib.h>
 
+#include "blas.h"
 #include "parser.h"
 #include "assert.h"
 #include "activations.h"
@@ -147,7 +148,10 @@ convolutional_layer parse_convolutional(list *options, size_params params)
     int n = option_find_int(options, "filters",1);
     int size = option_find_int(options, "size",1);
     int stride = option_find_int(options, "stride",1);
-    int pad = option_find_int(options, "pad",0);
+    int pad = option_find_int_quiet(options, "pad",0);
+    int padding = option_find_int_quiet(options, "padding",0);
+    if(pad) padding = size/2;
+
     char *activation_s = option_find_str(options, "activation", "logistic");
     ACTIVATION activation = get_activation(activation_s);
 
@@ -161,7 +165,7 @@ convolutional_layer parse_convolutional(list *options, size_params params)
     int binary = option_find_int_quiet(options, "binary", 0);
     int xnor = option_find_int_quiet(options, "xnor", 0);
 
-    convolutional_layer layer = make_convolutional_layer(batch,h,w,c,n,size,stride,pad,activation, batch_normalize, binary, xnor);
+    convolutional_layer layer = make_convolutional_layer(batch,h,w,c,n,size,stride,padding,activation, batch_normalize, binary, xnor);
     layer.flipped = option_find_int_quiet(options, "flipped", 0);
     layer.dot = option_find_float_quiet(options, "dot", 0);
 
@@ -234,9 +238,16 @@ layer parse_region(list *options, size_params params)
     int coords = option_find_int(options, "coords", 4);
     int classes = option_find_int(options, "classes", 20);
     int num = option_find_int(options, "num", 1);
+
+    params.w = option_find_int(options, "side", params.w);
+    params.h = option_find_int(options, "side", params.h);
+
     layer l = make_region_layer(params.batch, params.w, params.h, num, classes, coords);
     assert(l.outputs == params.inputs);
 
+    l.log = option_find_int_quiet(options, "log", 0);
+    l.sqrt = option_find_int_quiet(options, "sqrt", 0);
+
     l.softmax = option_find_int(options, "softmax", 0);
     l.max_boxes = option_find_int_quiet(options, "max",30);
     l.jitter = option_find_float(options, "jitter", .2);
@@ -278,6 +289,7 @@ cost_layer parse_cost(list *options, size_params params)
     COST_TYPE type = get_cost_type(type_s);
     float scale = option_find_float_quiet(options, "scale",1);
     cost_layer layer = make_cost_layer(params.batch, params.inputs, type, scale);
+    layer.ratio =  option_find_float_quiet(options, "ratio",0);
     return layer;
 }
 
@@ -324,6 +336,7 @@ maxpool_layer parse_maxpool(list *options, size_params params)
 {
     int stride = option_find_int(options, "stride",1);
     int size = option_find_int(options, "size",stride);
+    int padding = option_find_int_quiet(options, "padding", (size-1)/2);
 
     int batch,h,w,c;
     h = params.h;
@@ -332,7 +345,7 @@ maxpool_layer parse_maxpool(list *options, size_params params)
     batch=params.batch;
     if(!(h && w && c)) error("Layer before maxpool layer must output image.");
 
-    maxpool_layer layer = make_maxpool_layer(batch,h,w,c,size,stride);
+    maxpool_layer layer = make_maxpool_layer(batch,h,w,c,size,stride,padding);
     return layer;
 }
 
@@ -486,6 +499,7 @@ void parse_net_options(list *options, network *net)
     net->angle = option_find_float_quiet(options, "angle", 0);
     net->saturation = option_find_float_quiet(options, "saturation", 1);
     net->exposure = option_find_float_quiet(options, "exposure", 1);
+    net->hue = option_find_float_quiet(options, "hue", 0);
 
     if(!net->inputs && !(net->h && net->w && net->c)) error("No input parameters supplied");
 
@@ -1085,6 +1099,7 @@ void load_convolutional_weights(layer l, FILE *fp)
         fread(l.rolling_variance, sizeof(float), l.n, fp);
     }
     fread(l.filters, sizeof(float), num, fp);
+    //if(l.c == 3) scal_cpu(num, 1./256, l.filters, 1);
     if (l.flipped) {
         transpose_matrix(l.filters, l.c*l.size*l.size, l.n);
     }

src/region_layer.c

@@ -80,8 +80,8 @@ box get_region_box(float *x, int index, int i, int j, int w, int h, int adjust,
         b.w = logistic_activate(x[index + 2]);
         b.h = logistic_activate(x[index + 3]);
     }
-    if(adjust && b.w < .01) b.w = .01;
-    if(adjust && b.h < .01) b.h = .01;
+    //if(adjust && b.w < .01) b.w = .01;
+    //if(adjust && b.h < .01) b.h = .01;
     return b;
 }
 
@@ -149,6 +149,7 @@ void forward_region_layer(const region_layer l, network_state state)
                     l.delta[index + 4] = l.noobject_scale * ((0 - l.output[index + 4]) * logistic_gradient(l.output[index + 4]));
                     if(best_iou > .5) l.delta[index + 4] = 0;
 
+                    /*
                     if(*(state.net.seen) < 6400){
                         box truth = {0};
                         truth.x = (i + .5)/l.w;
@@ -157,6 +158,7 @@ void forward_region_layer(const region_layer l, network_state state)
                         truth.h = .5;
                         delta_region_box(truth, l.output, index, i, j, l.w, l.h, l.delta, LOG, 1);
                     }
+                    */
                 }
             }
         }

src/tag.c

@@ -45,6 +45,11 @@ void train_tag(char *cfgfile, char *weightfile, int clear)
     args.d = &buffer;
     args.type = TAG_DATA;
 
+    args.angle = net.angle;
+    args.exposure = net.exposure;
+    args.saturation = net.saturation;
+    args.hue = net.hue;
+
     fprintf(stderr, "%d classes\n", net.outputs);
 
     load_thread = load_data_in_thread(args);

src/utils.c

@@ -585,6 +585,13 @@ float rand_uniform(float min, float max)
     return ((float)rand()/RAND_MAX * (max - min)) + min;
 }
 
+float rand_scale(float s)
+{
+    float scale = rand_uniform(1, s);
+    if(rand()%2) return scale;
+    return 1./scale;
+}
+
 float **one_hot_encode(float *a, int n, int k)
 {
     int i;

src/utils.h

@@ -42,6 +42,7 @@ float mse_array(float *a, int n);
 float rand_normal();
 size_t rand_size_t();
 float rand_uniform(float min, float max);
+float rand_scale(float s);
 int rand_int(int min, int max);
 float sum_array(float *a, int n);
 float mean_array(float *a, int n);

src/voxel.c

@@ -9,9 +9,9 @@
 
 void extract_voxel(char *lfile, char *rfile, char *prefix)
 {
+#ifdef OPENCV
     int w = 1920;
     int h = 1080;
-#ifdef OPENCV
     int shift = 0;
     int count = 0;
     CvCapture *lcap = cvCaptureFromFile(lfile);
@@ -39,7 +39,7 @@ void extract_voxel(char *lfile, char *rfile, char *prefix)
     }
 
 #else
-printf("need OpenCV for extraction\n");
+    printf("need OpenCV for extraction\n");
 #endif
 }
 

src/yolo.c

@@ -54,6 +54,11 @@ void train_yolo(char *cfgfile, char *weightfile)
     args.d = &buffer;
     args.type = REGION_DATA;
 
+    args.angle = net.angle;
+    args.exposure = net.exposure;
+    args.saturation = net.saturation;
+    args.hue = net.hue;
+
     pthread_t load_thread = load_data_in_thread(args);
     clock_t time;
     //while(i*imgs < N*120){