Commit Graph

138 Commits

Author SHA1 Message Date
cb94ef9308 Use dont_show in the obj.data file to disable detection for a specific class 2018-12-29 21:52:31 +03:00
dc827f4c1c Minor fix for mAP calculation during training 2018-12-27 00:30:05 +03:00
49189c22e8 Fix mAP chart 2018-12-19 01:54:18 +03:00
f0bed8ef73 Fixed drawing mAP 2018-12-12 23:20:44 +03:00
dc7f8a32ae mAP calculation during training, if is used flag -map 2018-12-09 18:18:47 +03:00
21a4ec9390 Saving loss-chart for each 100 iterations automatically 2018-11-26 11:11:56 +03:00
9f7d7c58b5 Minor fixes. Use CUDA 10.0 2018-11-17 02:48:46 +03:00
bdbbb2fd69 Minor fix 2018-10-18 17:30:49 +03:00
acf1489981 Fixed detector test for 1-channel images (black-white / infrared) 2018-10-18 16:01:29 +03:00
7ee4135910 Fix OpenCV compilation. maxpool_layer on GPU uses cuDNN. 2018-10-10 13:51:36 +03:00
c02216120b Compile fix 2018-09-28 20:43:47 +03:00
5cc2a7e20d Added mistakes checking 2018-09-28 20:27:30 +03:00
a284a7da8d Try to use avx_hs() - slow and requires alignment 4096 bits < (l.size*l.size*l.c)
May be faster only from 8192 bits and more.
2018-08-08 19:08:58 +03:00
5fc1e9a918 ext_output: flush stdout after printing output
this allows tooling to watch for changes while the application is running.
2018-07-10 17:13:00 -05:00
cfc5fedbb6 Just used spaces for indents instead of Tabs 2018-07-10 23:29:15 +03:00
455b2fc06f Able letterbox for video 2018-06-21 14:45:43 +03:00
7f0079dec5 Output video-file has the same FPS as source video-stream. Exit from demo by ESC. 2018-06-19 15:00:36 +03:00
e1e5abe191 Fixed some security issues 2018-06-16 01:27:11 +03:00
22796c5a49 Minor fix 2018-06-10 13:58:37 +03:00
1c05ebf522 Minor fix 2018-06-07 03:39:30 +03:00
5d8e3526ed Fix for unreliable IP-cameras 2018-06-03 15:21:48 +03:00
01bec65708 "channel" parameter of [net] is used in detector when preparing images for net. 2018-05-30 19:51:55 +03:00
1b7a1a0f33 Fixed top-coordinate output when -ext_output is used for video 2018-05-28 19:02:05 +03:00
160eddddc4 Minor fix 2018-05-27 16:53:42 +03:00
f52040f0bf Classifier uses C++ OpenCV cv::VideoCapture that hasn't bugs. 2018-05-23 18:43:51 +03:00
a2666d5b3c Fixed replace_image_to_label(): replaces directories only for Pascal VOC and COCO, replaces only extensions for other. 2018-05-22 18:23:48 +03:00
a720bf74e2 Added flag -ext_output for detector demo. Fixed proportional resize of network for random=1 2018-05-22 00:15:04 +03:00
7b7c669f49 Added flag -save_labels to save detections as training labels txt-files (pseudo labeling concept) 2018-05-20 00:39:03 +03:00
3ebcc647b6 Fixed network resizing (random=1) for non-square networks 2018-05-14 12:20:38 +03:00
cda8171feb Merge pull request #765 from HotChick91/AlexeyAB-code
Exit with nonzero status on error
2018-05-10 01:38:02 +03:00
e96401774b Fixed compile error 2018-05-08 19:09:43 +03:00
fb56f6d569 Fixed README.md 2018-05-08 18:02:23 +03:00
0948df52b8 Merge pull request #741 from IlyaOvodov/Fix_detector_output
Output improvements for detector results:
2018-05-08 17:27:45 +03:00
1b5afb4583 Output improvements for detector results:
When printing detector results, output was done in random order, obfuscating results for interpreting. Now:
1. Text output includes coordinates of rects in (left,right,top,bottom in pixels) along with label and score
2. Text output is sorted by rect lefts to simplify finding appropriate rects on image
3. If several class probs are > thresh for some detection, the most probable is written first and coordinates for others are not repeated
4. Rects are imprinted in image in order by their best class prob, so most probable rects are always on top and not overlayed by less probable ones
5. Most probable label for rect is always written first
Also:
6. Message about low GPU memory include required amount

Upd:
* struct detection changes reverted to keep it unchanged
* cmd line key "-ext_output" is added to manage extended output (rect coords in detector test mode)

# Conflicts:
#	src/detector.c
2018-05-08 14:09:35 +03:00
ae4ec1a06d Exit with nonzero status on error 2018-05-07 22:11:57 +02:00
89354d0a0c Fixed memory leaks. And fixes for Web-camera and IP-camera. 2018-05-04 23:52:05 +03:00
4232ace376 Fix webcam memory leak 2018-05-04 18:37:33 +03:00
028696bf15 Output improvements for detector results:
When printing detector results, output was done in random order, obfuscating results for interpreting. Now:
1. Text output includes coordinates of rects in (left,right,top,bottom in pixels) along with label and score
2. Text output is sorted by rect lefts to simplify finding appropriate rects on image
3. If several class probs are > thresh for some detection, the most probable is written first and coordinates for others are not repeated
4. Rects are imprinted in image in order by their best class prob, so most probable rects are always on top and not overlayed by less probable ones
5. Most probable label for rect is always written first
Also:
6. Message about low GPU memory include required amount
2018-05-03 16:33:46 +03:00
92b25a7200 FIX #736: Detector does not draw rects for images with small resolutions 2018-05-03 12:43:48 +03:00
a0e288f0dc Increased speed of darknet.py. Fixed functions related to resize_image(). 2018-05-01 02:17:49 +03:00
c52fa47428 Loss-graph store automatically (iterations == max_batches) at the end of training 2018-04-16 13:09:10 +03:00
d75854c0fd Minor fix 2018-04-12 19:31:36 +03:00
d9ae3dd681 Added Yolo v3 2018-03-28 02:59:03 +03:00
fe4f01b1d8 Fixed uses of WebCam for OpenCV 3.x 2018-03-20 22:10:46 +03:00
537d135feb Improve training performance - batch-norm using cuDNN. 2018-03-20 02:16:51 +03:00
df076653e0 Minor fix 2018-03-13 19:04:39 +03:00
f0bc634a74 Fixed saving video file 2018-03-05 13:25:27 +03:00
c123f320fc Show avg-loss chart during Training, if compiled with OpenCV. Use -dont_show to disable. 2018-03-04 23:41:23 +03:00
5cd147031b Added param -dont_show to detect objects on servers without display, like Amazon EC2 2018-03-01 16:05:39 +03:00
35a66b73a2 Changed class to class_id 2018-03-01 15:14:58 +03:00