mirror of https://github.com/WongKinYiu/yolov7.git
detect.py: support of txt file list
FL with configured weighted loss : alpha test.py: tSNE over the cropped bbox featurespull/2071/head
hanoch
parent
30c9907884
commit
6b8d33fa31
|
|
@ -1,8 +1,8 @@
|
|||
lr0: 0.01 #0.001 # initial learning rate (SGD=1E-2, Adam=1E-3)
|
||||
lr0: 0.005 #0.001 # initial learning rate (SGD=1E-2, Adam=1E-3)
|
||||
lrf: 0.01 # final OneCycleLR learning rate (lr0 * lrf)
|
||||
momentum: 0.937 # SGD momentum/Adam beta1
|
||||
weight_decay: 0.005 # optimizer weight decay 5e-4 It resolve mAP of overfitting test
|
||||
warmup_epochs: 0.0 # warmup epochs (fractions ok)
|
||||
warmup_epochs: 3.0 # warmup epochs (fractions ok)
|
||||
warmup_momentum: 0.8 # warmup initial momentum
|
||||
warmup_bias_lr: 0.001 #0.001 # warmup initial bias lr
|
||||
loss_ota: 1 #1 # use ComputeLossOTA, use 0 for faster training
|
||||
|
|
|
|||
|
|
@ -0,0 +1,18 @@
|
|||
# COCO 2017 dataset http://cocodataset.org
|
||||
|
||||
# download command/URL (optional)
|
||||
#Make symbolic link
|
||||
# sudo ln -s ~hanoch/projects/tir_frames_rois /mnt/Data/hanoch/tir_frames_rois
|
||||
path: /mnt/Data/hanoch/tir_frames_rois/yolo7_tir_data_all #/home/hanoch/projects/tir_frames_rois/tir_car_44person_31 #/home/hanochk/tir_frames_rois/yolo7_tir_data
|
||||
# train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/]
|
||||
train: ./yolov7/tir_od/training_set_yolov7941_w_center_roi_no_swiss_no_38B_20b_22c_n_png.txt #training_set_yolov7941_w_center_roi_no_swiss_no_38B_20b_22c_w_png.txt
|
||||
val: ./yolov7/tir_od/tir_tiff_w_center_roi_validation_set.txt #./yolov7/tir_od/tir_tiff_car_person_min_size_44_31_validation_set.txt #./yolov7/tir_od/validation_set.txt #./yolov7/tir_od/val_tir_od.txt #./yolov7/tir_od/validation_set.txt # 5000 images
|
||||
#test: ./tir_od/test-dev2017.txt # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794
|
||||
|
||||
# number of classes
|
||||
nc: 2
|
||||
|
||||
# class names
|
||||
names: ['car', 'person']
|
||||
|
||||
|
||||
|
|
@ -0,0 +1,18 @@
|
|||
# COCO 2017 dataset http://cocodataset.org
|
||||
|
||||
# download command/URL (optional)
|
||||
#Make symbolic link
|
||||
# sudo ln -s ~hanoch/projects/tir_frames_rois /mnt/Data/hanoch/tir_frames_rois
|
||||
path: /mnt/Data/hanoch/tir_frames_rois/yolo7_tir_data_all #/home/hanoch/projects/tir_frames_rois/tir_car_44person_31 #/home/hanochk/tir_frames_rois/yolo7_tir_data
|
||||
# train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/]
|
||||
train: ./yolov7/tir_od/training_set_yolov7941_w_center_roi_no_swiss_no_38B_20b_22c_n_png_train_3_cls_jan_25.txt # training_set_yolov7941_w_center_roi_no_swiss_no_38B_20b_22c_n_png.txt #./yolov7/tir_od/training_set_yolov7941_w_center_roi.txt
|
||||
val: ./yolov7/tir_od/tir_tiff_w_center_roi_validation_set_train_cls_usa.txt #tir_tiff_w_center_roi_validation_set.txt #./yolov7/tir_od/tir_tiff_car_person_min_size_44_31_validation_set.txt #./yolov7/tir_od/validation_set.txt #./yolov7/tir_od/val_tir_od.txt #./yolov7/tir_od/validation_set.txt # 5000 images
|
||||
#test: ./tir_od/test-dev2017.txt # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794
|
||||
|
||||
# number of classes
|
||||
nc: 3
|
||||
|
||||
# class names
|
||||
names: ['car', 'person', 'train']
|
||||
|
||||
|
||||
|
|
@ -0,0 +1,18 @@
|
|||
# COCO 2017 dataset http://cocodataset.org
|
||||
|
||||
# download command/URL (optional)
|
||||
|
||||
path: /mnt/Data/hanoch/tir_frames_rois/yolo7_tir_data_all #/home/hanoch/projects/tir_frames_rois/tir_tiff_tiff_files #/home/hanoch/projects/tir_frames_rois/tir_car_44person_31 #/home/hanochk/tir_frames_rois/yolo7_tir_data
|
||||
# train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/]
|
||||
#train: ./yolov7/tir_od/training_set.txt #./yolov7/tir_od/training_set.txt # ./yolov7/tir_od/tir_tiff_car_person_min_size_44_31_training_set.txt #./yolov7/tir_od/training_set.txt #./yolov7/tir_od/training_set.txt # 118287 images
|
||||
#val: ./yolov7/tir_od/validation_set.txt #./yolov7/tir_od/tir_tiff_car_person_min_size_44_31_validation_set.txt #./yolov7/tir_od/validation_set.txt #./yolov7/tir_od/val_tir_od.txt #./yolov7/tir_od/validation_set.txt # 5000 images
|
||||
|
||||
test: ./yolov7/tir_od/fog_data_set.txt #Test51a_Test40A_test_set.txt #Test51a_Test40A_test_set_tiny.txt #Test51a_Test40A_test_set.txt # data at tir_tiff_tiff_files
|
||||
#test: ./yolov7/tir_od/test_set/tir_tiff_tiff_folder_test_set.txt # data at tir_tiff_tiff_files
|
||||
#test: ./tir_od/test-dev2017.txt # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794
|
||||
|
||||
# number of classes
|
||||
nc: 2
|
||||
|
||||
# class names
|
||||
names: ['car', 'person']
|
||||
|
|
@ -5,14 +5,14 @@
|
|||
# sudo ln -s ~hanoch/projects/tir_frames_rois /mnt/Data/hanoch/tir_frames_rois
|
||||
path: /mnt/Data/hanoch/tir_frames_rois/yolo7_tir_data_all #/home/hanoch/projects/tir_frames_rois/tir_car_44person_31 #/home/hanochk/tir_frames_rois/yolo7_tir_data
|
||||
# train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/]
|
||||
train: ./yolov7/tir_od/training_set_overfit.txt #./yolov7/tir_od/training_set_yolov7941_w_center_roi.txt
|
||||
val: ./yolov7/tir_od/training_set_overfit.txt #./yolov7/tir_od/tir_tiff_car_person_min_size_44_31_validation_set.txt #./yolov7/tir_od/validation_set.txt #./yolov7/tir_od/val_tir_od.txt #./yolov7/tir_od/validation_set.txt # 5000 images
|
||||
train: ./yolov7/tir_od/training_set_train_only_overfit.txt #training_set_overfit.txt #./yolov7/tir_od/training_set_yolov7941_w_center_roi.txt
|
||||
val: ./yolov7/tir_od/training_set_train_only_overfit.txt #training_set_overfit.txt #./yolov7/tir_od/tir_tiff_car_person_min_size_44_31_validation_set.txt #./yolov7/tir_od/validation_set.txt #./yolov7/tir_od/val_tir_od.txt #./yolov7/tir_od/validation_set.txt # 5000 images
|
||||
#test: ./tir_od/test-dev2017.txt # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794
|
||||
|
||||
# number of classes
|
||||
nc: 2
|
||||
nc: 3
|
||||
|
||||
# class names
|
||||
names: ['car', 'person']
|
||||
names: ['car', 'person', 'train']
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -0,0 +1,18 @@
|
|||
# COCO 2017 dataset http://cocodataset.org
|
||||
|
||||
# download command/URL (optional)
|
||||
#Make symbolic link
|
||||
# sudo ln -s ~hanoch/projects/tir_frames_rois /mnt/Data/hanoch/tir_frames_rois
|
||||
path: /mnt/Data/hanoch/tir_frames_rois/yolo7_tir_data_all #/home/hanoch/projects/tir_frames_rois/tir_car_44person_31 #/home/hanochk/tir_frames_rois/yolo7_tir_data
|
||||
# train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/]
|
||||
train: ./yolov7/tir_od/training_set_overfit.txt #./yolov7/tir_od/training_set_yolov7941_w_center_roi.txt
|
||||
val: ./yolov7/tir_od/training_set_overfit.txt #./yolov7/tir_od/tir_tiff_car_person_min_size_44_31_validation_set.txt #./yolov7/tir_od/validation_set.txt #./yolov7/tir_od/val_tir_od.txt #./yolov7/tir_od/validation_set.txt # 5000 images
|
||||
#test: ./tir_od/test-dev2017.txt # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794
|
||||
|
||||
# number of classes
|
||||
nc: 3
|
||||
|
||||
# class names
|
||||
names: ['car', 'person', 'Train']
|
||||
|
||||
|
||||
|
|
@ -7,7 +7,7 @@ path: /mnt/Data/hanoch/tir_frames_rois/yolo7_tir_data_all #/home/hanoch/projects
|
|||
#train: ./yolov7/tir_od/training_set.txt #./yolov7/tir_od/training_set.txt # ./yolov7/tir_od/tir_tiff_car_person_min_size_44_31_training_set.txt #./yolov7/tir_od/training_set.txt #./yolov7/tir_od/training_set.txt # 118287 images
|
||||
#val: ./yolov7/tir_od/validation_set.txt #./yolov7/tir_od/tir_tiff_car_person_min_size_44_31_validation_set.txt #./yolov7/tir_od/validation_set.txt #./yolov7/tir_od/val_tir_od.txt #./yolov7/tir_od/validation_set.txt # 5000 images
|
||||
|
||||
test: ./yolov7/tir_od/test_set/Test51a_Test40A_test_set.txt # data at tir_tiff_tiff_files
|
||||
test: ./yolov7/tir_od/test_set/Test51a_Test40A_test_set.txt #Test51a_Test40A_test_set_tiny.txt #Test51a_Test40A_test_set.txt # data at tir_tiff_tiff_files
|
||||
#test: ./yolov7/tir_od/test_set/tir_tiff_tiff_folder_test_set.txt # data at tir_tiff_tiff_files
|
||||
#test: ./tir_od/test-dev2017.txt # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794
|
||||
|
||||
|
|
|
|||
|
|
@ -0,0 +1,18 @@
|
|||
# COCO 2017 dataset http://cocodataset.org
|
||||
|
||||
# download command/URL (optional)
|
||||
|
||||
path: /mnt/Data/hanoch/tir_frames_rois/yolo7_tir_data_all #/home/hanoch/projects/tir_frames_rois/tir_tiff_tiff_files #/home/hanoch/projects/tir_frames_rois/tir_car_44person_31 #/home/hanochk/tir_frames_rois/yolo7_tir_data
|
||||
# train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/]
|
||||
#train: ./yolov7/tir_od/training_set.txt #./yolov7/tir_od/training_set.txt # ./yolov7/tir_od/tir_tiff_car_person_min_size_44_31_training_set.txt #./yolov7/tir_od/training_set.txt #./yolov7/tir_od/training_set.txt # 118287 images
|
||||
#val: ./yolov7/tir_od/validation_set.txt #./yolov7/tir_od/tir_tiff_car_person_min_size_44_31_validation_set.txt #./yolov7/tir_od/validation_set.txt #./yolov7/tir_od/val_tir_od.txt #./yolov7/tir_od/validation_set.txt # 5000 images
|
||||
|
||||
test: ./yolov7/tir_od/test_set/Test51a_Test40A_test_set.txt #Test51a_Test40A_test_set_tiny.txt #Test51a_Test40A_test_set.txt # data at tir_tiff_tiff_files
|
||||
#test: ./yolov7/tir_od/test_set/tir_tiff_tiff_folder_test_set.txt # data at tir_tiff_tiff_files
|
||||
#test: ./tir_od/test-dev2017.txt # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794
|
||||
|
||||
# number of classes
|
||||
nc: 3
|
||||
|
||||
# class names
|
||||
names: ['car', 'person', 'train']
|
||||
14
detect.py
14
detect.py
|
|
@ -21,10 +21,10 @@ from utils.torch_utils import select_device, load_classifier, time_synchronized,
|
|||
|
||||
def detect(save_img=False):
|
||||
source, weights, view_img, save_txt, imgsz, trace = opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size, not opt.no_trace
|
||||
save_img = not opt.nosave and not source.endswith('.txt') # save inference images
|
||||
webcam = source.isnumeric() or source.endswith('.txt') or source.lower().startswith(
|
||||
save_img = not opt.nosave #and not source.endswith('.txt') # save inference images
|
||||
webcam = source.isnumeric() or source.lower().startswith( # removed HK or source.endswith('.txt')
|
||||
('rtsp://', 'rtmp://', 'http://', 'https://'))
|
||||
|
||||
webcam = webcam and not source.endswith('.txt')
|
||||
# Directories
|
||||
save_dir = Path(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok)) # increment run
|
||||
(save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True) # make dir
|
||||
|
|
@ -61,7 +61,8 @@ def detect(save_img=False):
|
|||
dataset = LoadImages(source, img_size=imgsz, stride=stride,
|
||||
scaling_type=opt.norm_type, input_channels=opt.input_channels,
|
||||
no_tir_signal=opt.no_tir_signal,
|
||||
tir_channel_expansion=opt.tir_channel_expansion)
|
||||
tir_channel_expansion=opt.tir_channel_expansion,
|
||||
rel_path_for_list_files=opt.rel_path_for_list_files)
|
||||
|
||||
# Get names and colors
|
||||
names = model.module.names if hasattr(model, 'module') else model.names
|
||||
|
|
@ -218,6 +219,9 @@ if __name__ == '__main__':
|
|||
|
||||
parser.add_argument('--save-path', default='', help='save to project/name')
|
||||
|
||||
parser.add_argument('--rel-path-for-list-files', default='/mnt/Data/hanoch/tir_frames_rois/yolo7_tir_data_all', help='')
|
||||
|
||||
|
||||
|
||||
opt = parser.parse_args()
|
||||
|
||||
|
|
@ -229,7 +233,6 @@ if __name__ == '__main__':
|
|||
|
||||
print(opt)
|
||||
#check_requirements(exclude=('pycocotools', 'thop'))
|
||||
|
||||
with torch.no_grad():
|
||||
if opt.update: # update all models (to fix SourceChangeWarning)
|
||||
for opt.weights in ['yolov7.pt']:
|
||||
|
|
@ -237,6 +240,7 @@ if __name__ == '__main__':
|
|||
strip_optimizer(opt.weights)
|
||||
else:
|
||||
detect()
|
||||
# --source yolov7/tir_od/fog_data_set.txt
|
||||
|
||||
"""
|
||||
python detect.py --weights yolov7.pt --conf 0.25 --img-size 640 --source inference/images/horses.jpg
|
||||
|
|
|
|||
45
test.py
45
test.py
|
|
@ -17,7 +17,7 @@ from utils.metrics import ap_per_class, ConfusionMatrix, range_bar_plot, range_p
|
|||
from utils.plots import plot_images, output_to_target, plot_study_txt, append_to_txt
|
||||
from utils.torch_utils import select_device, time_synchronized, TracedModel
|
||||
import pandas as pd
|
||||
|
||||
from yolo_object_embeddings import ObjectEmbeddingVisualizer
|
||||
|
||||
def object_size_to_range(obj_height_pixels: float, focal:int, class_id:int=1):
|
||||
class_height = {0:1.5, 1:1.8} # car Sedan height = 1.5 m , person height is 1.8m
|
||||
|
|
@ -104,7 +104,7 @@ def test(data,
|
|||
log_imgs = min(wandb_logger.log_imgs, 100)
|
||||
# Dataloader
|
||||
|
||||
|
||||
embed_analyse = kwargs.get('embed_analyse', False)
|
||||
if not training:
|
||||
if device.type != 'cpu':
|
||||
model(torch.zeros(1, opt.input_channels, imgsz, imgsz).to(device).type_as(next(model.parameters()))) # run once
|
||||
|
|
@ -123,12 +123,21 @@ def test(data,
|
|||
dataloader = create_dataloader(data[task], imgsz, batch_size, gs, opt, hyp, pad=0.5, augment=False, rect=False, #rect was True # HK@@@ TODO : why pad =0.5?? only effective in rect=True in test time ? https://github.com/ultralytics/ultralytics/issues/13271
|
||||
prefix=colorstr(f'{task}: '), rel_path_images=data['path'], num_cls=data['nc'])[0]
|
||||
|
||||
labels = np.concatenate(dataloader.dataset.labels, 0)
|
||||
class_labels = torch.tensor(labels[:, 0]) # classes
|
||||
|
||||
|
||||
if v5_metric:
|
||||
print("Testing with YOLOv5 AP metric...")
|
||||
|
||||
seen = 0
|
||||
confusion_matrix = ConfusionMatrix(nc=nc, conf=conf_thres, iou_thres=iou_thres) # HK per conf per iou_thresh
|
||||
names = {k: v for k, v in enumerate(model.names if hasattr(model, 'names') else model.module.names)}
|
||||
|
||||
if not training:
|
||||
print(100 * '==')
|
||||
print('Test set labels {} count : {}'.format(names, torch.bincount(class_labels.long(), minlength=nc) + 1))
|
||||
|
||||
coco91class = coco80_to_coco91_class()
|
||||
s = ('%20s' + '%12s' * 6) % ('Class', 'Images', 'Labels', 'P', 'R', 'mAP@.5', 'mAP@.5:.95')
|
||||
p, r, f1, mp, mr, map50, map, t0, t1 = 0., 0., 0., 0., 0., 0., 0., 0., 0.
|
||||
|
|
@ -145,8 +154,12 @@ def test(data,
|
|||
'score'
|
||||
])
|
||||
|
||||
stats_all_large, stats_person_medium = [], []
|
||||
if embed_analyse:
|
||||
obj_embed_viz = ObjectEmbeddingVisualizer(model=model, device=device)
|
||||
features_acm = torch.empty((0, 1024)) # embedding dim of last scale 1024x20x20
|
||||
labels_acm = np.array([])
|
||||
|
||||
stats_all_large, stats_person_medium = [], []
|
||||
if dataloader.dataset.use_csv_meta_data_file:
|
||||
|
||||
n_bins_of100m = 20
|
||||
|
|
@ -209,8 +222,16 @@ def test(data,
|
|||
# out = non_max_suppression(out, conf_thres=conf_thres, iou_thres=iou_thres, labels=lb, multi_label=False) # Does thresholding for class : list of detections, on (n,6) tensor per image [xyxy, conf, cls]
|
||||
t1 += time_synchronized() - t
|
||||
|
||||
if trace and embed_analyse and np.sum([x.numel() for x in out])>0: # features are being saved/clone in the trace model version only TODO for others
|
||||
features, labels = obj_embed_viz.extract_object_grounded_features(feature_maps=model.features,
|
||||
predictions=out,
|
||||
image_shape=img.shape)
|
||||
features_acm = torch.cat((features_acm, features.detach().cpu()), dim=0)
|
||||
labels_acm = np.concatenate((labels_acm, labels), axis=0)
|
||||
|
||||
# Statistics per image
|
||||
for si, pred in enumerate(out): # [bbox_coors, objectness_logit, class]
|
||||
|
||||
labels = targets[targets[:, 0] == si, 1:]
|
||||
nl = len(labels)
|
||||
tcls = labels[:, 0].tolist() if nl else [] # target class
|
||||
|
|
@ -422,6 +443,11 @@ def test(data,
|
|||
|
||||
# Compute statistics
|
||||
stats = [np.concatenate(x, 0) for x in zip(*stats)] # to numpy
|
||||
if trace and embed_analyse:
|
||||
embeddings = obj_embed_viz.visualize_object_embeddings(features_acm,
|
||||
labels_acm,
|
||||
path=save_dir,
|
||||
tag=opt.conf_thres)
|
||||
|
||||
if not training or 1:
|
||||
stats_person_medium = [np.concatenate(x, 0) for x in zip(*stats_person_medium)] # to numpy
|
||||
|
|
@ -782,6 +808,9 @@ if __name__ == '__main__':
|
|||
|
||||
parser.add_argument('--csv-metadata-path', default='', help='save to project/name')
|
||||
|
||||
parser.add_argument('--embed-analyse', action='store_true', help='')
|
||||
|
||||
|
||||
opt = parser.parse_args()
|
||||
|
||||
if opt.tir_channel_expansion: # operates over 3 channels
|
||||
|
|
@ -812,7 +841,8 @@ if __name__ == '__main__':
|
|||
save_conf=opt.save_conf,
|
||||
trace=not opt.no_trace,
|
||||
v5_metric=opt.v5_metric,
|
||||
hyp=hyp)
|
||||
hyp=hyp,
|
||||
embed_analyse=opt.embed_analyse)
|
||||
|
||||
elif opt.task == 'speed': # speed benchmarks
|
||||
for w in opt.weights:
|
||||
|
|
@ -867,9 +897,12 @@ mAP:
|
|||
Fixed wether csv P/R
|
||||
--weights /mnt/Data/hanoch/runs/train/yolov7999/weights/best.pt --device 0 --batch-size 16 --data data/tir_od_test_set.yaml --img-size 640 --verbose --norm-type single_image_percentile_0_1 --input-channels 1 --project test --task test --csv-metadata-path tir_od/tir_tiff_seq_png_3_class_fixed_whether.xlsx --iou-thres 0.6 --conf 0.65
|
||||
|
||||
|
||||
FOG
|
||||
--weights /mnt/Data/hanoch/runs/train/yolov7999/weights/best.pt --device 0 --batch-size 16 --data data/tir_od_fog_set.yaml --img-size 640 --verbose --norm-type single_image_percentile_0_1 --input-channels 1 --project test --task test --csv-metadata-path tir_od/tir_tiff_seq_png_3_class_fixed_whether.xlsx --conf 0.01 --iou-thres 0.6
|
||||
--weights /mnt/Data/hanoch/runs/train/yolov7999/weights/best.pt --device 0 --batch-size 16 --data data/tir_od_fog_set.yaml --img-size 640 --verbose --norm-type single_image_percentile_0_1 --input-channels 1 --project test --task test --csv-metadata-path tir_od/tir_tiff_seq_png_3_class_fixed_whether.xlsx --conf 0.65 --iou-thres 0.6
|
||||
|
||||
Locomotive
|
||||
|
||||
--weights /mnt/Data/hanoch/runs/train/yolov71107/weights/best.pt --device 0 --batch-size 16 --data data/tir_od_test_set_3_class_train.yaml --img-size 640 --verbose --norm-type single_image_percentile_0_1 --input-channels 1 --project test --task test --iou-thres 0.6 --conf 0.1 --embed-analyse
|
||||
------- Error analysis ------------
|
||||
1st run with conf_th=0.0001 then observe the desired threshold, re-run with the desired threshold abd observe images with bboxes given the deired threshold
|
||||
"""
|
||||
File diff suppressed because it is too large
Load Diff
|
|
@ -1056,7 +1056,6 @@
|
|||
./TIR5_V60_DEC21_Test51A_ML_RD_IL_2021_11_07_14_27_00_FS210_XGA_1951_2510_LIEL_left_roi_210_2479.tiff
|
||||
./TIR5_V60_DEC21_Test51A_ML_RD_IL_2021_11_07_14_27_00_FS210_XGA_1951_2510_LIEL_right_roi_210_2479.tiff
|
||||
./TIR5_V60_DEC21_Test51A_ML_RD_IL_2021_11_07_14_27_00_FS210_XGA_1951_2510_LIEL_left_roi_210_2480.tiff
|
||||
./TIR5_V60_DEC21_Test51A_ML_RD_IL_2021_11_07_14_27_00_FS210_XGA_1951_2510_LIEL_right_roi_210_2480.tiff
|
||||
./TIR5_V60_DEC21_Test51A_ML_RD_IL_2021_11_07_14_27_00_FS210_XGA_1951_2510_LIEL_left_roi_210_2481.tiff
|
||||
./TIR5_V60_DEC21_Test51A_ML_RD_IL_2021_11_07_14_27_00_FS210_XGA_1951_2510_LIEL_left_roi_210_2482.tiff
|
||||
./TIR5_V60_DEC21_Test51A_ML_RD_IL_2021_11_07_14_27_00_FS210_XGA_1951_2510_LIEL_left_roi_210_2483.tiff
|
||||
|
|
@ -1190,23 +1189,14 @@
|
|||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8348.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_right_roi_210_8348.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8351.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_right_roi_210_8351.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8355.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_right_roi_210_8355.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8358.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_right_roi_210_8358.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8361.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_right_roi_210_8361.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8364.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_right_roi_210_8364.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8367.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_right_roi_210_8367.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8371.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_right_roi_210_8371.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8375.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_right_roi_210_8375.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8379.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_right_roi_210_8379.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8384.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8389.tiff
|
||||
./TIR1_v60_JAN22_TEST51A_ML_RD_IL_2021_11_04_03_25_33_FS_210_XGA_8229_8982_Shaked_left_roi_210_8394.tiff
|
||||
|
|
|
|||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
42
train.py
42
train.py
|
|
@ -3,6 +3,7 @@ import logging
|
|||
import math
|
||||
import os
|
||||
import random
|
||||
import re
|
||||
import time
|
||||
from copy import deepcopy
|
||||
from pathlib import Path
|
||||
|
|
@ -20,7 +21,7 @@ from torch.cuda import amp
|
|||
from torch.nn.parallel import DistributedDataParallel as DDP
|
||||
from torch.utils.tensorboard import SummaryWriter
|
||||
from tqdm import tqdm
|
||||
import re
|
||||
|
||||
import test # import test.py to get mAP after each epoch
|
||||
|
||||
try:
|
||||
|
|
@ -40,25 +41,26 @@ from models.yolo import Model
|
|||
from utils.autoanchor import check_anchors
|
||||
from utils.datasets import create_dataloader
|
||||
from utils.general import labels_to_class_weights, increment_path, labels_to_image_weights, init_seeds, \
|
||||
fitness, strip_optimizer, get_latest_run, check_dataset, check_file, check_git_status, check_img_size, \
|
||||
check_requirements, print_mutation, set_logging, one_cycle, colorstr
|
||||
fitness, strip_optimizer, get_latest_run, check_dataset, check_file, check_img_size, \
|
||||
print_mutation, set_logging, one_cycle, colorstr
|
||||
from utils.google_utils import attempt_download
|
||||
from utils.loss import ComputeLoss, ComputeLossOTA
|
||||
from utils.plots import plot_images, plot_labels, plot_results, plot_evolution
|
||||
from utils.plots import plot_images, plot_results, plot_evolution
|
||||
from utils.torch_utils import ModelEMA, select_device, intersect_dicts, torch_distributed_zero_first, is_parallel
|
||||
from utils.wandb_logging.wandb_utils import WandbLogger, check_wandb_resume
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
clear_ml = True
|
||||
|
||||
from clearml import Task, Logger
|
||||
from clearml import Task
|
||||
|
||||
if clear_ml: # clearml support
|
||||
|
||||
task = Task.init(
|
||||
project_name="TIR_OD",
|
||||
task_name="train yolov7 with dummy test" # output_uri = True model torch.save will uploaded to file server or =/mnt/myfolder or AWS or Azure
|
||||
)
|
||||
task_name="train yolov7 'train' class" # output_uri = True model torch.save will uploaded to file server or =/mnt/myfolder or AWS or Azure
|
||||
# output_uri='azure://company.blob.core.windows.net/folder'
|
||||
)
|
||||
# Task.execute_remotely() will invoke the job immidiately over the remote and not DeV
|
||||
task.set_base_docker(docker_image="nvcr.io/nvidia/pytorch:24.09-py3", docker_arguments="--shm-size 8G")
|
||||
# clear_ml can capture graph like tensorboard
|
||||
|
|
@ -413,6 +415,7 @@ def train(hyp, opt, device, tb_writer=None):
|
|||
model.hyp = hyp # attach hyperparameters to model
|
||||
model.gr = 1.0 # iou loss ratio (obj_loss = 1.0 or iou)
|
||||
model.class_weights = labels_to_class_weights(dataset.labels, nc).to(device) * nc # attach class weights
|
||||
class_inverse_freq = labels_to_class_weights(dataset.labels, nc).to(device)
|
||||
model.names = names
|
||||
|
||||
# Start training
|
||||
|
|
@ -430,8 +433,14 @@ def train(hyp, opt, device, tb_writer=None):
|
|||
else:
|
||||
scaler = torch.amp.GradScaler("cuda", enabled=opt.amp) if is_torch_240 else torch.cuda.amp.GradScaler(enabled=opt.amp)
|
||||
|
||||
compute_loss_ota = ComputeLossOTA(model) # init loss class
|
||||
compute_loss = ComputeLoss(model) # init loss class
|
||||
loss_weight = torch.tensor([])
|
||||
if opt.loss_weight:
|
||||
loss_weight = class_inverse_freq
|
||||
if 'loss_ota' not in hyp or hyp['loss_ota'] == 1:
|
||||
compute_loss_ota = ComputeLossOTA(model, loss_weight=loss_weight) # init loss class
|
||||
else:
|
||||
compute_loss = ComputeLoss(model, loss_weight=loss_weight) # init loss class
|
||||
|
||||
logger.info(f'Image sizes {imgsz} train, {imgsz_test} test\n'
|
||||
f'Using {dataloader.num_workers} dataloader workers\n'
|
||||
f'Logging results to {save_dir}\n'
|
||||
|
|
@ -448,6 +457,8 @@ def train(hyp, opt, device, tb_writer=None):
|
|||
if 0: # HK TODO remove later The anomaly mode tells you about the nan. If you remove this and you have the nan error again, you should have an additional stack trace that tells you about the forward function (make sure to enable the anomaly mode before the you run the forward).
|
||||
torch.autograd.set_detect_anomaly(True)
|
||||
|
||||
print(100 * '==')
|
||||
print('Training set labels {} count : {}'.format(names, torch.bincount(c.long(), minlength=nc) + 1))
|
||||
for epoch in range(start_epoch, epochs): # epoch ------------------------------------------------------------------
|
||||
model.train()
|
||||
|
||||
|
|
@ -479,8 +490,10 @@ def train(hyp, opt, device, tb_writer=None):
|
|||
optimizer.zero_grad()
|
||||
|
||||
for i, (imgs, targets, paths, _) in pbar: # batch -------------------------------------------------------------
|
||||
# print(np.unique(targets, return_counts=True))
|
||||
# print(np.bincount(targets[:,1].long(), minlength=nc))
|
||||
ni = i + nb * epoch # number integrated batches (since train start) i.e. iterations
|
||||
# imgs = imgs.to(device, non_blocking=True).float() / 255.0 # uint8 to float32, 0-255 to 0.0-1.0 @@HK TODO is that standartization ?
|
||||
|
||||
imgs = imgs.to(device, non_blocking=True).float()
|
||||
|
||||
# Warmup
|
||||
|
|
@ -504,7 +517,7 @@ def train(hyp, opt, device, tb_writer=None):
|
|||
|
||||
with amp.autocast(enabled=cuda): # to decrease GPU VRAM turn off OTA loss see what happen HT TODO ::
|
||||
# with amp.autocast(enabled=(cuda and opt.amp)):
|
||||
pred = model(imgs) # forward
|
||||
pred = model(imgs) # forward [B, C, W,H, [bbox[4], objectness[1], class-conf[nc]]]
|
||||
if 'loss_ota' not in hyp or hyp['loss_ota'] == 1:
|
||||
loss, loss_items = compute_loss_ota(pred, targets.to(device), imgs) # loss scaled by batch_size
|
||||
else:
|
||||
|
|
@ -761,6 +774,10 @@ if __name__ == '__main__':
|
|||
|
||||
parser.add_argument('--csv-metadata-path', default='', help='save to project/name')
|
||||
|
||||
parser.add_argument('--loss-weight', action='store_true', help='weight the loss by 1/freq to compensate for imbalanced data')
|
||||
|
||||
parser.add_argument('--embed-analyse', action='store_true', help='')
|
||||
|
||||
|
||||
opt = parser.parse_args()
|
||||
# Only for clearML env
|
||||
|
|
@ -966,7 +983,8 @@ Overfit 640x640
|
|||
tir_od_overfit.yaml
|
||||
--workers 8 --device 0 --batch-size 32 --data data/tir_od_overfit.yaml --img-size 640 --weights /mnt/Data/hanoch/tir_frames_rois/yolov7.pt --cfg cfg/training/yolov7.yaml --name yolov7 --hyp hyp.tir_od_v7_overfit.yaml --adam --norm-type single_image_percentile_0_1 --input-channels 1 --linear-lr --epochs 100 --nosave --gamma-aug-prob 0.2 --cache-images
|
||||
|
||||
|
||||
# 3 classe renew yolov7999 list
|
||||
--workers 8 --device 0 --batch-size 24 --data data/tir_od_center_roi_aug_list_train_cls.yaml --img 640 640 --weights /mnt/Data/hanoch/tir_frames_rois/yolov7.pt --cfg cfg/training/yolov7.yaml --name yolov7 --hyp hyp.tir_od.tiny_aug_gamma_scaling_before_mosaic_rnd_scaling_no_ota.yaml --adam --norm-type single_image_percentile_0_1 --input-channels 1 --linear-lr --epochs 100 --gamma-aug-prob 0.1 --cache-images --image-weights
|
||||
#########################################################
|
||||
Extended model for higher resolution YOLO7E6
|
||||
# --workers 8 --device 0 --batch-size 8 --data data/tir_od_center_roi_aug_list_full_res.yaml --weights /mnt/Data/hanoch/tir_frames_rois/yolov7-e6.pt --img-size [768, 1024] --cfg cfg/deploy/yolov7-e6.yaml --name yolov7e --hyp hyp.tir_od.aug_gamma_scaling_before_mosaic_rnd_scaling_e6_full_res.yaml --adam --norm-type single_image_percentile_0_1 --input-channels 1 --linear-lr --epochs 2 --gamma-aug-prob 0.3 --cache-images --rect
|
||||
|
|
|
|||
|
|
@ -237,7 +237,8 @@ class _RepeatSampler(object):
|
|||
class LoadImages: # for inference
|
||||
def __init__(self, path, img_size=640, stride=32,
|
||||
scaling_type='standardization', img_percentile_removal=0.3, beta=0.3, input_channels=3,
|
||||
tir_channel_expansion=False, no_tir_signal=False):
|
||||
tir_channel_expansion=False, no_tir_signal=False,
|
||||
rel_path_for_list_files=''):
|
||||
|
||||
p = str(Path(path).absolute()) # os-agnostic absolute path
|
||||
if '*' in p:
|
||||
|
|
@ -245,7 +246,11 @@ class LoadImages: # for inference
|
|||
elif os.path.isdir(p):
|
||||
files = sorted(glob.glob(os.path.join(p, '*.*'))) # dir
|
||||
elif os.path.isfile(p):
|
||||
files = [p] # files
|
||||
if path.endswith('.txt'):
|
||||
files = self.parse_image_file_names(path, rel_path_for_list_files)
|
||||
|
||||
else:
|
||||
files = [p] # files
|
||||
else:
|
||||
raise Exception(f'ERROR: {p} does not exist')
|
||||
|
||||
|
|
@ -273,6 +278,36 @@ class LoadImages: # for inference
|
|||
self.tir_channel_expansion = tir_channel_expansion
|
||||
self.is_tir_signal = not (no_tir_signal)
|
||||
|
||||
def parse_image_file_names(self, path, rel_path_for_list_files):
|
||||
try:
|
||||
f = [] # image files
|
||||
for p in path if isinstance(path, list) else [path]:
|
||||
p = Path(p) # os-agnostic
|
||||
if p.is_dir(): # dir
|
||||
f += glob.glob(str(p / '**' / '*.*'), recursive=True)
|
||||
# f = list(p.rglob('**/*.*')) # pathlib
|
||||
elif p.is_file(): # file
|
||||
with open(p, 'r') as t:
|
||||
t = t.read().strip().splitlines()
|
||||
parent = str(p.parent) + os.sep
|
||||
if bool(rel_path_for_list_files):
|
||||
f += [os.path.join(rel_path_for_list_files, x.replace('./', '')).rstrip() if x.startswith(
|
||||
'./') else x for x
|
||||
in t] # local to global path
|
||||
else:
|
||||
f += [x.replace('./', parent).rstrip() if x.startswith('./') else x for x in
|
||||
t] # local to global path
|
||||
|
||||
# f += [p.parent / x.lstrip(os.sep) for x in t] # local to global path (pathlib)
|
||||
else:
|
||||
raise Exception(f'{p} does not exist')
|
||||
self.img_files = sorted(
|
||||
[x.replace('/', os.sep) for x in f if x.split('.')[-1].lower() in img_formats])
|
||||
# self.img_files = sorted([x for x in f if x.suffix[1:].lower() in img_formats]) # pathlib
|
||||
assert self.img_files, f' No images found'
|
||||
except Exception as e:
|
||||
raise Exception(f'Error loading data from {path}: {e}\nSee {help_url}')
|
||||
return f
|
||||
|
||||
def __iter__(self):
|
||||
self.count = 0
|
||||
|
|
@ -677,7 +712,7 @@ class LoadImagesAndLabels(Dataset): # for training/testing
|
|||
if self.use_csv_meta_data_file:
|
||||
df = load_csv_xls_2_df(self.csv_meta_data_file)
|
||||
self.df_metadata = pd.DataFrame(columns=['sensor_type', 'part_in_day', 'weather_condition', 'country', 'train_state', 'tir_frame_image_file_name'])
|
||||
# TODO :HK @@ itereate tqdm(zip(self.img_files, self.label_files) and upon --force-csv-list remove missing entries in the csv from train/test lists!!!
|
||||
# TODO :HK @@ itereate tqdm(zip(self.img_files, self.label_files) and upon --force-csv-list remove missing entries from the csv in train/test lists!!!
|
||||
for ix, fname in enumerate(self.img_files):
|
||||
file_name = fname.split('/')[-1]
|
||||
if not (df['tir_frame_image_file_name'] == file_name).any():
|
||||
|
|
|
|||
|
|
@ -147,7 +147,17 @@ class FocalLoss(nn.Module):
|
|||
else: # 'none'
|
||||
return loss
|
||||
|
||||
"""
|
||||
https://github.com/ultralytics/ultralytics/issues/10406
|
||||
Define a list of alpha values:
|
||||
|
||||
alpha_list = [0.25, 0.5, 0.75] # Example alpha values for three classes
|
||||
Modify the focal loss calculation in your model’s loss function:
|
||||
|
||||
alpha_factor = torch.tensor([alpha_list[i] for i in labels]).to(device)
|
||||
alpha_factor = alpha_factor * labels + (1 - alpha_factor) * (1 - labels)
|
||||
loss *= alpha_factor
|
||||
"""
|
||||
class QFocalLoss(nn.Module):
|
||||
# Wraps Quality focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)
|
||||
def __init__(self, loss_fcn, gamma=1.5, alpha=0.25):
|
||||
|
|
@ -421,7 +431,7 @@ class APLoss(torch.autograd.Function):
|
|||
# Dual obj and cls losses and outputs inherited from Joseph Redmon's original YOLOv3
|
||||
class ComputeLoss:
|
||||
# Compute losses
|
||||
def __init__(self, model, autobalance=False):
|
||||
def __init__(self, model, autobalance=False, loss_weight=torch.tensor([])):
|
||||
super(ComputeLoss, self).__init__()
|
||||
device = next(model.parameters()).device # get model device
|
||||
h = model.hyp # hyperparameters
|
||||
|
|
@ -429,14 +439,17 @@ class ComputeLoss:
|
|||
# Define criteria
|
||||
BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['cls_pw']], device=device))
|
||||
BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['obj_pw']], device=device))
|
||||
|
||||
self.loss_weight = loss_weight
|
||||
# Class label smoothing https://arxiv.org/pdf/1902.04103.pdf eqn 3
|
||||
self.cp, self.cn = smooth_BCE(eps=h.get('label_smoothing', 0.0)) # positive, negative BCE targets
|
||||
|
||||
# Focal loss
|
||||
g = h['fl_gamma'] # focal loss gamma
|
||||
if g > 0:
|
||||
BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g)
|
||||
alpha = 0.25 # default by base code
|
||||
if loss_weight.numel()>0:
|
||||
alpha = loss_weight # Overide the default from the paper
|
||||
BCEcls, BCEobj = FocalLoss(BCEcls, g, alpha=alpha), FocalLoss(BCEobj, g)
|
||||
|
||||
det = model.module.model[-1] if is_parallel(model) else model.model[-1] # Detect() module
|
||||
self.balance = {3: [4.0, 1.0, 0.4]}.get(det.nl, [4.0, 1.0, 0.25, 0.06, .02]) # P3-P7
|
||||
|
|
@ -555,11 +568,12 @@ class ComputeLoss:
|
|||
|
||||
class ComputeLossOTA:
|
||||
# Compute losses
|
||||
def __init__(self, model, autobalance=False):
|
||||
def __init__(self, model, autobalance=False, loss_weight=torch.tensor([])):
|
||||
super(ComputeLossOTA, self).__init__()
|
||||
device = next(model.parameters()).device # get model device
|
||||
h = model.hyp # hyperparameters
|
||||
|
||||
if loss_weight.numel() > 0:
|
||||
raise ValueError('Not imp. yet')
|
||||
# Define criteria
|
||||
BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['cls_pw']], device=device))
|
||||
BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['obj_pw']], device=device))
|
||||
|
|
|
|||
|
|
@ -370,6 +370,18 @@ class TracedModel(nn.Module):
|
|||
print(" model is traced! \n")
|
||||
|
||||
def forward(self, x, augment=False, profile=False):
|
||||
out = self.model(x)
|
||||
out = self.model(x)# feat = torch.mean(out[-1], dim=[2, 3])
|
||||
self.features = [x.clone().detach() for x in out]
|
||||
out = self.detect_layer(out)
|
||||
return out
|
||||
return out
|
||||
# https://stackoverflow.com/questions/75319661/how-to-extract-and-visualize-feature-value-for-an-arbitrary-layer-during-inferen
|
||||
|
||||
"""
|
||||
def plot_ts_feature_maps(feature_maps, tag=''):
|
||||
import matplotlib
|
||||
feature_maps = feature_maps.to(torch.float32)
|
||||
ts.show(feature_maps)
|
||||
ts.save(feature_maps, '/home/hanoch/projects/tir_od/output/' + tag + '.jpg')
|
||||
|
||||
plot_ts_feature_maps(torch.mean(out[-1], dim=[2, 3]))
|
||||
"""
|
||||
Loading…
Reference in New Issue