OrgLion-ML
/
darknet
mirror of https://github.com/AlexeyAB/darknet.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
1,605 Commits 1 Branch 7 Tags 361 MiB
Commit Graph

86 Commits

Author SHA1 Message Date
AlexeyAB 9bd88d7fd7 weighted [shortcut] layer 2020-01-08 02:20:42 +03:00
AlexeyAB b267e34487 multilayer [shortcut] in progress 2020-01-06 23:27:41 +03:00
AlexeyAB 489ea7fdce Minor fix 2020-01-01 21:38:06 +03:00
AlexeyAB 2c08a20b22 Added flag -benchmark_layers for: classifier demo, detector test, detector train 2019-12-28 23:03:30 +03:00
AlexeyAB 114a7f942b Added command line param -dontuse_opencv for training Classifier. Also use GaussianBlur instead of bilateralFilter for blur=1 for training Classifier and Detector. 2019-12-28 18:46:10 +03:00
AlexeyAB af54d95184 Added stretch_sway 2019-12-22 16:00:20 +03:00
AlexeyAB 005714a10c Added [local_avgpool] layer 2019-12-20 20:28:35 +03:00
AlexeyAB e33ecb785e Added param counters_per_class=100,200,... <for each class> to [yolo] and [Gaussian_yolo] to solve class-imbalance-problem 2019-12-17 18:03:42 +03:00
AlexeyAB 1626f9d495 Some fixes 2019-12-16 17:53:58 +03:00
AlexeyAB 5e035565ff Rotate conv-kernel 2019-12-14 18:11:09 +03:00
AlexeyAB a08c872564 Sway conv-kernel 2019-12-14 03:07:01 +03:00
AlexeyAB 642c065c0e Fixed DropBlock [dropout] dropblock_size_abs=7 dropblock_size_rel=0.5 2019-12-13 01:22:03 +03:00
AlexeyAB 1df3ddc7d6 Added DropBlock. Use [dropout] dropblock=1 dropblock_size=0.5 probability=0.1 2019-12-11 13:16:41 +03:00
AlexeyAB 2a873f3448 Added label_smooth_eps=0.1 for [net] layer for Label Smoothing for Classifier 2019-12-08 17:25:01 +03:00
AlexeyAB 318919e1cb Added label_smooth_eps=0.1 for [yolo] and [Gaussian_yolo] layer for Label Smoothing 2019-12-08 16:39:28 +03:00
AlexeyAB c9c745ccf1 Added [convolutional] activation=normalize_channels_softmax 2019-12-06 23:25:12 +03:00
AlexeyAB 68db8efb27 Added Fusion features for ASFF and for BiFPN(EfficientDet) 2019-11-28 20:34:01 +03:00
AlexeyAB 0d98f20be1 You can train with mini_batch=128 (batch=256 subdivisions=2) on GPU 8 GB VRAM + 128 GB CPU-RAM and more. Set in cfg-file: optimized_memory=3 and workspace_size_limit_MB=2000 or 4000 2019-11-26 23:18:04 +03:00
AlexeyAB 48914023da Added init_cpu() function to init CPU just to speedup the 1st detection (if GPU=0) 2019-11-25 18:10:39 +03:00
AlexeyAB 8cb3ee4e79 diounms_sort() fixed 2019-11-23 18:59:29 +03:00
AlexeyAB 4f70fc14e9 Added DIoU and CIoU to [yolo] and [Gaussian_yolo] 2019-11-23 16:42:43 +03:00
AlexeyAB 6e5bdf1282 Bug fix 2019-11-16 18:16:37 +03:00
AlexeyAB e7e85b358c Added delta_yolo_box_accumulate(). Added iou_thresh=0.213 parameter to [yolo] layer IoU(anchor,truth). 2019-11-15 01:14:55 +03:00
AlexeyAB ded620ac6b fixed memory deallocation 2019-11-12 15:12:06 +03:00
AlexeyAB d91d59a22f Optimized memory allocation for Detection (inference only), without allocation memory for training 2019-11-09 22:48:18 +03:00
AlexeyAB 0fa9c8f105 Added groups= and groupd_id= params to the [route] layer 2019-11-08 01:06:49 +03:00
AlexeyAB bf8ea4183d Added MISH activation, use activation=mish in [convolutional] layers 2019-11-07 01:15:58 +03:00
AlexeyAB 52e3bb252f Use non-blocking sockets for JSON_server and MJPEG_server 2019-11-04 14:39:20 +03:00
AlexeyAB b3a2495298 Added Gaussian YOLOv3 layer [Gaussian_yolo] 2019-10-25 20:47:17 +03:00
AlexeyAB be5d0d6693 Added assisted_excitation=1 for [convolutional] layer on GPU 2019-09-03 01:35:05 +03:00
AlexeyAB 11b8b2bddf Added antialiasing=1 param for [convolutional]-layer on GPU 2019-09-02 02:13:11 +03:00
AlexeyAB 8d80a65288 Max pool layer can use stride=2 or stride_x=2 stride_y=4 (isn't tested well) 2019-08-27 14:04:39 +03:00
AlexeyAB c1f184249e Added assisted_excitation_forward for training on CPU 2019-08-08 00:29:06 +03:00
AlexeyAB 2fa9792c56 Added [sam] layer 2019-08-02 19:29:05 +03:00
Stefano Sinigardi b9e7cd3d35 introduce functions to detect build features 2019-07-18 10:48:19 +02:00
AlexeyAB 54e2d0b0e8 Added ability to use letter_box resizing in darknet.py sample 2019-07-01 19:07:13 +03:00
AlexeyAB d4402d29c2 Added maxpool_depth= and out_channels= params to [maxpool] 2019-06-30 13:53:11 +03:00
AlexeyAB 88cccfcad4 Added Swish-activation 2019-06-26 15:48:22 +03:00
AlexeyAB c9129c2078 Added letter_box=1 param in [net] section (cfg-file) for keeping aspect ratio during training 2019-06-22 17:04:00 +03:00
AlexeyAB cc41339805 Added [scale_channels] layer for squeeze-and-excitation blocks 2019-06-19 23:36:02 +03:00
AlexeyAB bd689f91ac Added TridentNet (state-of-art on MS COCO) 2019-06-13 14:43:23 +03:00
AlexeyAB 88ce9dcca6 Added dilation parameter for convolutional, conv_lstm and crnn layers 2019-06-08 19:41:44 +03:00
AlexeyAB 57baf1a487 Added Mixup data augmentation if is used mixup=1 2019-06-06 02:55:11 +03:00
Alexey 2347913ef1
Merge pull request #2952 from cenit/dev/cenit/include 
[headers] remove many installed files which should remain private, add more fixes for downstream projects
 2019-06-05 17:37:28 +03:00
AlexeyAB 1f6f2e1162 Compile fix 2019-06-04 21:46:11 +03:00
AlexeyAB eac26226a7 Added scale_x_y param to [yolo]-layer (for sigmoid). 
In previous commit is used GIoU(generalized-iou) from  https://github.com/generalized-iou/g-darknet
 2019-06-04 21:20:59 +03:00
AlexeyAB 6e13527f06 Added GIoU 2019-06-01 01:27:18 +03:00
Stefano Sinigardi a3d5cc4664
Merge branch 'master' into dev/cenit/include 2019-05-25 20:27:19 +02:00
AlexeyAB 9c9344a1ff Added flag -letter_box for Detection 2019-05-21 23:47:47 +03:00
AlexeyAB 038289eb7d Some conv-lstm, sgdr and other fixes 2019-05-14 14:35:22 +03:00