mirrors/yolov7
1
0
Fork 0
mirror of https://github.com/WongKinYiu/yolov7.git synced 2025-06-03 21:54:57 +08:00
Code Issues Projects Releases Wiki Activity
202 Commits 9 Branches 1 Tag
Commit Graph

202 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
hanoch3
0a12c34633 Remove center_crop /s that had samll truncated objects 
+missing png/s and other tiff from the final annotation list : tir_tiff_seq_png_3_class_fixed_whether_copied_dataset_label.xlsx
 2025-03-24 16:24:29 +02:00
hanoch3
2cc7e05cc8 Update validation set : Test36B 13/11 
training-set Test36B 12/11
+missing png/s and other tiff from the final annotation list : tir_tiff_seq_png_3_class_fixed_whether_copied_dataset_label.xlsx
 2025-03-24 14:57:46 +02:00
hanoch3
c90fa54f1e Update validation set : Test36B 13/11 
training-set Test36B 12/11
+missing png/s and other tiff from the final annotation list : tir_tiff_seq_png_3_class_fixed_whether_copied_dataset_label.xlsx
 2025-03-24 14:57:24 +02:00
hanoch3
4fdfef3ffe add missing png files to the training set training_set_yolov7941_w_center_roi_no_swiss_no_38B_20b_22c_n_png_train_3_cls_jan_25.txt 2025-03-23 17:45:04 +02:00
hanoch3
ab6b4bd341 BCE loss with weight vector - fix 
Trigger when AP fall
 2025-03-23 16:05:37 +02:00
hanoch
8766685f94 Fix image_data_creator.py : truncated objects in lower part of image: fix effective height computation 
datasets.py : Remove empty annotated image entries
weights file moved to HDD
 2025-02-27 12:45:47 +02:00
hanoch
52320239d4 ONNX reorder output labels No. by pre-NMs processing, model name over PR 2025-02-20 10:38:08 +02:00
hanoch
78cd0cb8f6 detect.py: support of txt file list 
FL with configured weighted loss : alpha
test.py: tSNE over the cropped bbox features
 2025-02-16 10:05:06 +02:00
hanoch
6b8d33fa31 detect.py: support of txt file list 
FL with configured weighted loss : alpha
test.py: tSNE over the cropped bbox features
 2025-02-16 10:03:45 +02:00
hanoch
30c9907884 gignore add xml 2025-02-03 15:20:20 +02:00
hanoch
d00653795f True prediction equal the same range cell as GT. Fix bug where no preds but GT exist. whether statistics, ONNX save pre-processed file, 2025-02-03 15:14:16 +02:00
hanoch
4c0190e554 Test51a_Test40A_test_set.txt : remove truncated/small objects on the right crop. Whether condition based breakdown 2025-01-30 13:34:19 +02:00
hanoch
b8e71e931a adding debug filename of each prediction/GT. Legend to P/R per bin @ threshold 2025-01-29 15:46:14 +02:00
hanoch
f4266d2161 adding debug filename of each prediction/GT. Legend to P/R per bin @ threshold 2025-01-29 15:35:29 +02:00
hanoch
e88bd2fd10 Fix bug : np.array.size() condition, pre range iteration, clear all range p/r stats, height based range, p/r per range per class 2025-01-28 15:00:05 +02:00
hanoch
59b7be6b91 Fix bug 2025-01-27 14:03:51 +02:00
hanoch
085ee89bfc mAP breakdown sensor/time in day/ mAP per range : Dataset : add csv with metadata of each image, caching labels omit label w/ class no not in the list. range_bar plot, add plot class support. Onnx RT: adding noise to dataset, dataset with ext. noise 2025-01-27 10:10:38 +02:00
hanoch
0f8f0772fe Onnx for Prediction w/o GT 2025-01-13 15:44:41 +02:00
hanoch
973e43446d Onnx for Prediction w/o GT 2025-01-13 15:40:19 +02:00
hanoch
ab1f39e24a Overfit 640 settings commit before checkout old 2025-01-08 11:27:05 +02:00
hanoch
b8a2a10ffe Union train list mp4/tiff/seq 2025-01-07 17:05:07 +02:00
hanoch
772cee9084 datasets.py: 768x1024, random padding, roi based scaling . still open issue the prediction over the random padding area should be omitted from calculation 2025-01-06 17:28:32 +02:00
hanoch
221429a81b datasets.py: 768x1024, random padding, roi based scaling . still open issue the prediction over the random padding area should be omitted from calculation 2025-01-06 17:26:49 +02:00
hanoch
76e6315e99 Onnx model P/R curve, fixed, 1st IOU criterion per image than mAP overall. 2024-12-24 12:32:20 +02:00
hanoch
54a0cc226b NMS-ORT built in Onnx model P/R curve fits theory. 2024-12-23 17:03:39 +02:00
hanoch
ee65c332a3 Fix mAP over Onnx wrapper for old TIR model 2024-12-16 18:04:06 +02:00
hanoch
9aec3afb56 adapt to TIR env export.py. Fix threshold computation for Precision recall. test.py : adding preds/gts per image to csv for error analysis 2024-12-16 17:02:01 +02:00
hanoch
5d99ec8b2c Onnx model OnnxRT for eval old TIR model, validate annotations over PNG 2024-12-11 17:28:56 +02:00
hanoch
abe52501fe comments 2024-12-05 17:17:47 +02:00
hanoch
6421910e14 center_roi 2024-12-03 17:02:27 +02:00
hanoch
542cc99ada PNG/MP$ dataset __getitem__() adaptation. Fix paste_in loading many patches by avoiding normalizing the PNG 2024-12-01 17:37:59 +02:00
hanoch
705aa5ea87 Fix random padding disabled on mosaic that ovverides ROI 2024-11-26 14:53:41 +02:00
hanoch
6ab8732fbe hyp effective in ClearML, center_roi train/val/test list 2024-11-26 12:05:31 +02:00
hanoch
13a313c649 Gamma augmentation robustness to Random padding, clip 0<image<1, new training set excluding Test19G swiss with bad NUC compensation 2024-11-26 09:51:51 +02:00
hanoch
c6e82c42f6 Random padding- post processing w/ all augmentations permutations.standartization Stability increasing epsilon to 1e-5. Random padding. Replace Albumentation Gamma aug by skimage one more stable 2024-11-25 15:51:22 +02:00
hanoch
6c577e95c9 New Train val set split Swiss-snow-2018 to training(test19g_sy_rd_ch) val:Test46c_ml_rd_il 2024-11-19 13:35:43 +02:00
hanoch
2de1a2593c Random padding, Val/train new split append to previous list 2024-11-17 15:10:26 +02:00
hanoch
4afc52b1cd Filter validation set from RF leakage USA. Add tag per PR curve 2024-11-13 16:49:31 +02:00
hanoch
7391d5fc94 Clean bad recordings from France 2024-11-11 11:22:01 +02:00
hanoch
ddeff5a8fc increased training/val set 24500/5346 2024-11-10 17:36:16 +02:00
hanoch
5e5dfde1f1 Fix rnd perspective fillng the padded pixels by TIR mean value to reduce edges artifacts. 
mAP per med/large exclusively
 2024-11-06 16:54:53 +02:00
hanoch
b2bc3773a1 hyp : scaling and mosaic w/ scliang before mosaic 2024-11-04 10:13:35 +02:00
hanoch
15755c31b9 dataset.py: align test() loader to train 
scaling before mosaic with predefined filling_value according to img precision chnaged by scaling before mosaic. filling uint16 with image.mean()
detect.py : align to latest changes as test.py
test.py : support in large object size metrices stats_all_large
plots.py : plot detections with th>0.1 (not 0.25)
 2024-11-04 10:08:35 +02:00
hanoch
c040cf8265 enable random perspective 2024-10-31 11:13:08 +02:00
hanoch
e16bd3ffad random seed, hyp for random perspective 2024-10-30 15:43:20 +02:00
hanoch
5639572615 plot PR curve thresholds and R@P 
remove no_aug
 2024-10-27 14:49:26 +02:00
hanoch
e2f0c15a00 remove torch anomaly() 2024-10-22 10:32:51 +03:00
hanoch
f366ed0f30 stability of denum in data norm/standard 2024-10-21 15:57:07 +03:00
hanoch
5f8046d05f claer ml debug 2024-10-20 17:08:46 +03:00
hanoch
102d731395 claer ml debug 2024-10-20 16:51:29 +03:00