OrgLion-ML
/
darknet
mirror of https://github.com/AlexeyAB/darknet.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
2,268 Commits 1 Branch 7 Tags 361 MiB
Commit Graph

161 Commits

Author SHA1 Message Date
Muhammad Maaz 5a9970b24a
Add support for less or more frequent mAP calculation while training (#4918) 
* Feature: Add support to pass '-mAP_epochs' from command line.

'-mAP_epochs' decide after how many epochs mAP should be calculated. Default value is 4 which means the mAP will be calculated after each 4 epochs.
 2023-09-01 16:20:25 +02:00
Stefano Sinigardi 2f4ab0887c
Gl integration (#8833) 
* Start adding support for running darknet from OpenGL textures

* Get CUDA-GL interop working (at least without segfaults...)

* Copy the GL texture to the CUDA input buffer

* Remove printf statements

* Code cleanup

* Fix a function call

* Remove the #ifdef GPU guard in darknet.h

* Use the width and height from the network instead of hard-coding

* Install libgles2-mesa-dev for some ubuntu targets

* Get the build working on Windows

* Add an OpenGL-CUDA integration option for CMake

* Bump the version-date in vcpkg.json

* Remove ifdef in darknet.h for the network_predict_gl_texture() function

* Use target_compile_definitions to set CUDA_OPENGL_INTEGRATION flag

* Re-add the #ifdef CUDA_OPENGL_INTEGRATION guard

---------

Co-authored-by: Brian Schwind <brianmschwind@gmail.com>
 2023-08-26 03:50:00 +02:00
Tobit Flatscher d2c3788539
feat: Move set_batch_network to public API (#8782) 2023-08-02 08:54:08 +02:00
Sarim Mehdi 9ff8653d99
Made inference faster (this is especially useful when using Yolo9000) (#8009) 
* Update network.c

custom_get_region_detections function now keeps track of class index with the highest probability.

* Update darknet.h

Added best_class_idx to detection struct

* Update darknet.py

added python code for faster negative removal and also faster non-max suppression
 2021-08-25 18:10:21 +03:00
Sergey Nuzhny d669680879
Add batch inference on C++ (#7915) 
* Add batch inference on C++

* Return default params

* Add make_nms parameter
 2021-07-18 18:58:01 +03:00
Stéphane Charette 5e73447fa8
allocation and error functions now take a filename, function, and line number to better diagnose where problems happen (#7925) 2021-07-18 06:01:10 +03:00
8k 4361b7a322
add thresholds (#6831) 2021-07-16 00:28:25 +03:00
AlexeyAB c2221f07f8 Fixed adversarial training 2021-06-29 23:54:58 +03:00
AlexeyAB 846c79b6d4 Added [empty]/[silence] and [implicit] layers 2021-05-11 22:59:21 +03:00
AlexeyAB 53db0163eb Added CUDA graphs/streams. To speed up the inference use: [net] use_cuda_graph = 1 2021-02-28 08:57:44 +03:00
AlexeyAB 560fda3494 [net] weights_reject_freq=1001 2020-12-02 14:42:05 +03:00
AlexeyAB 80e1da22ee Added [net] ema_alpha=0.998 2020-11-29 21:30:41 +03:00
AlexeyAB 332ec6496e [net] equidistant_point=200000 num_sigmas_reject_badlabels=3 badlabels_rejection_percentage=0.1 2020-11-28 17:35:54 +03:00
AlexeyAB 333cc14a06 [net] equidistant_point=1 instead of LR-scheduler 2020-11-24 01:41:00 +03:00
AlexeyAB f056fc3b6a Speedup [yolo] layer 2020-11-23 23:32:55 +03:00
AlexeyAB 8c9c517189 [yolo] new_coords=1 2020-11-16 16:29:06 +03:00
AlexeyAB 8993a24747 Added: [net] contrastive_color=0 and [contrastive] contrastive_neg_max 2020-09-21 05:48:15 +03:00
AlexeyAB acd8f31969 Added [yolo] obj_normalizer=1.0 delta_normalizer=1.0 2020-08-30 01:55:49 +03:00
AlexeyAB 05dee78fa3 Added [convolutional] coordconv=1 for GPU-only 2020-08-17 02:13:46 +03:00
AlexeyAB 0d8718a434 Added layer: [history] history_size=4 2020-08-16 17:55:10 +03:00
AlexeyAB 10fdf9a852 P_constrastive_f_det_gpu() 2020-08-05 03:18:33 +03:00
AlexeyAB 0b36b29b6a Minor fix 2020-08-02 17:56:17 +03:00
AlexeyAB c61771e1f1 Merge branch 'master' of github.com:AlexeyAB/darknet 2020-07-25 14:57:00 +03:00
AlexeyAB 74f32a3928 Improved speed of [contrastive] layer. Added Hard-Mish activation. Added [maxpool] maxpool_zero_nonmax=1 param 2020-07-25 14:56:48 +03:00
Alexey c13db432bd
Merge pull request #4814 from Dev-Nash/master 
Minor addition
 2020-07-25 14:51:12 +03:00
Juuso Alasuutari f5258a83e2 Fix member initialization ordering 
Signed-off-by: Juuso Alasuutari <juuso.alasuutari@gmail.com>
 2020-07-10 15:10:09 +03:00
AlexeyAB 37753e0cb4 Improved tracking (new params) 2020-07-08 00:08:45 +03:00
AlexeyAB ccb392ddf2 Show track_id for models with contrastive learning 2020-07-07 19:30:49 +03:00
AlexeyAB ce2ae2c5c3 put embeddings to the detections 2020-07-07 02:49:43 +03:00
AlexeyAB 2fca3c399b Fixed bug in contrastive learning 2020-07-07 00:43:17 +03:00
AlexeyAB 73beba671d Contrastive loss is fixed to work with both Classifier and Detector 2020-07-06 21:25:23 +03:00
AlexeyAB f13532ce60 Contrastive learning for training Detector to track objects 2020-07-06 01:50:44 +03:00
AlexeyAB 8f900493c6 SAT for contrastive loss 2020-06-21 03:02:52 +03:00
AlexeyAB a553899505 Added Supervised and Unsupervised Contrastive loss for Classifier [net] contrastive=1 unsupervised=1 [contrastive] classes=1000 2020-06-15 19:22:47 +03:00
AlexeyAB 09991d0488 Added Contrastive loss 2020-06-14 05:11:22 +03:00
AlexeyAB 6c6f04a9b3 [net] mosaic_bound=1 fixed bboxes going out of bounds. Count rewritten bboxes. Changed self-adversarial-training. 2020-06-07 04:41:01 +03:00
AlexeyAB c625cc6503 Added [conv_lstm] time_normalizer=0.5 2020-06-01 17:51:29 +03:00
AlexeyAB e455cdef78 Added [conv_lstm] lstm_activation=tanh shortcut=1 2020-06-01 05:13:21 +03:00
AlexeyAB 269878a44d Added [conv_lstm] bottleneck=1 - to make the layer cheaper 2020-05-29 03:17:19 +03:00
AlexeyAB 73768ef6e3 Added [net] max_chart_loss=50 for scale y-axis of Loss-chart 2020-05-15 18:52:20 +03:00
AlexeyAB 88a39b31d3 Added param [yolo] resize=2.0 for resizing image [1/2 - 2x] without changing aspect ratio (while jitter=0.3 changes aspect ratio) 2020-05-10 02:17:00 +03:00
AlexeyAB f5105342be Added Gradient Centralization: [convolutional] grad_centr=1 2020-05-09 22:29:07 +03:00
AlexeyAB 424632f336 Fixed typo weights_normalizion -> weights_normalization 2020-05-08 13:20:57 +03:00
AlexeyAB ccafff912b Demo speed fixed 2020-05-08 12:31:08 +03:00
AlexeyAB 0fffb686fd Added param [yolo] objectness_smooth=1 2020-05-07 17:22:09 +03:00
AlexeyAB 6ebaef3095 Added GELU activation 2020-04-27 04:36:39 +03:00
AlexeyAB b573eab63f Minor fix 2020-04-22 20:49:27 +03:00
Alexey 88f28f7fcc
Merge pull request #4099 from enesozi/master 
Batch inference added
 2020-04-22 14:46:38 +03:00
AlexeyAB 4786d557f9 Fixed memory leak for Training with OPENCV=1 - create pool of CPU-threads only once (bug in OpenCV - it allocates memory for each thread without deallocation) 2020-04-18 19:01:20 +03:00
AlexeyAB d4b2ed9d22 Added param train_only_bn=1 to train only Batch-normalization for this and previous layers 2020-04-17 19:06:15 +03:00