Add support for image absolute path by using the "parent" directory in the data.yaml

cfg/training/yolov7-tiny.yaml

@@ -1,6 +1,6 @@
 # parameters
-nc: 80  # number of classes
-depth_multiple: 1.0  # model depth multiple
+nc: 2  # number of classes
+depth_multiple: 1.0  # model depth multiple  @@ HK TODO:
 width_multiple: 1.0  # layer channel multiple
 
 # anchors

data/coco_2_tir.yaml

@@ -0,0 +1,21 @@
+# COCO 2017 dataset http://cocodataset.org
+
+# download command/URL (optional)
+path: /home/hanoch/projects/tir_frames_rois/yolo7_tir_coco_classes_data_all #/home/hanoch/projects/tir_frames_rois/yolo7_tir_data_all   #/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/val_set_1_file.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/validation_set.txt #./yolov7/tir_od/val_tir_od.txt #./yolov7/tir_od/validation_set.txt  # 5000 images
+
+# number of classes
+nc: 80
+
+# class names
+names: [ 'person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat', 'traffic light',
+         'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow',
+         'elephant', 'bear', 'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee',
+         'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard',
+         'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple',
+         'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch',
+         'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 'cell phone',
+         'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 'teddy bear',
+         'hair drier', 'toothbrush' ]

data/hyp.scratch.tiny.yaml

@@ -1,4 +1,4 @@
-lr0: 0.01  # initial learning rate (SGD=1E-2, Adam=1E-3)
+lr0: 0.0005  # 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: 0.0005  # optimizer weight decay 5e-4
@@ -14,12 +14,12 @@ iou_t: 0.20  # IoU training threshold
 anchor_t: 4.0  # anchor-multiple threshold
 # anchors: 3  # anchors per output layer (0 to ignore)
 fl_gamma: 0.0  # focal loss gamma (efficientDet default gamma=1.5)
-hsv_h: 0.015  # image HSV-Hue augmentation (fraction)
-hsv_s: 0.7  # image HSV-Saturation augmentation (fraction)
-hsv_v: 0.4  # image HSV-Value augmentation (fraction)
+hsv_h: 0.0  # image HSV-Hue augmentation (fraction)
+hsv_s: 0.0  # image HSV-Saturation augmentation (fraction)
+hsv_v: 0.0  # image HSV-Value augmentation (fraction)
 degrees: 0.0  # image rotation (+/- deg)
-translate: 0.1  # image translation (+/- fraction)
-scale: 0.5  # image scale (+/- gain)
+translate: 0.0  # image translation (+/- fraction)
+scale: 0.0  # image scale (+/- gain)
 shear: 0.0  # image shear (+/- deg)
 perspective: 0.0  # image perspective (+/- fraction), range 0-0.001
 flipud: 0.0  # image flip up-down (probability)

data/hyp.tir_od.tiny.yaml

@@ -0,0 +1,36 @@
+lr0: 0.001 #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: 3.0  # warmup epochs (fractions ok)
+warmup_momentum: 0.8  # warmup initial momentum
+warmup_bias_lr: 0.001 #0.001  # warmup initial bias lr
+box: 0.05  # box loss gain
+cls: 0.5  # cls loss gain
+cls_pw: 1.0  # cls BCELoss positive_weight
+obj: 1.0  # obj loss gain (scale with pixels)
+obj_pw: 1.0  # obj BCELoss positive_weight
+iou_t: 0.20  # IoU training threshold
+anchor_t: 4.0  # anchor-multiple threshold
+anchors: 2  # anchors per output layer (0 to ignore)  @@HK was 3
+fl_gamma: 1.5 #1.5  # focal loss gamma (efficientDet default gamma=1.5)
+hsv_h: 0.0  # image HSV-Hue augmentation (fraction)
+hsv_s: 0.0  # image HSV-Saturation augmentation (fraction)
+hsv_v: 0.0  # image HSV-Value augmentation (fraction)
+degrees: 0.0  # image rotation (+/- deg)
+translate: 0.0  # image translation (+/- fraction)
+scale: 0.0  # image scale (+/- gain)
+shear: 0.0  # image shear (+/- deg)
+perspective: 0.0  # image perspective (+/- fraction), range 0-0.001
+flipud: 0.3  # image flip up-down (probability)
+fliplr: 0.5  # image flip left-right (probability)
+mosaic: 0.0  # image mosaic (probability)
+mixup: 0.0  # image mixup (probability)
+copy_paste: 0.0  # image copy paste (probability)
+paste_in: 0.0  # image copy paste (probability), use 0 for faster training : cutout
+loss_ota: 0 #1 # use ComputeLossOTA, use 0 for faster training
+inversion: 0.5 #opposite temperature
+drc_per_ch_percentile: 0.3 #[0, 0.2, 0.5]
+img_percentile_removal: 0.3
+beta : 0.3
+random_perspective : 0

data/hyp.tir_od.tiny_aug.yaml

@@ -0,0 +1,36 @@
+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: 3.0  # warmup epochs (fractions ok)
+warmup_momentum: 0.8  # warmup initial momentum
+warmup_bias_lr: 0.001 #0.001  # warmup initial bias lr
+box: 0.05  # box loss gain
+cls: 0.5  # cls loss gain
+cls_pw: 1.0  # cls BCELoss positive_weight
+obj: 1.0  # obj loss gain (scale with pixels)
+obj_pw: 1.0  # obj BCELoss positive_weight
+iou_t: 0.20  # IoU training threshold
+anchor_t: 4.0  # anchor-multiple threshold
+anchors: 2  # anchors per output layer (0 to ignore)  @@HK was 3
+fl_gamma: 1.5 #1.5  # focal loss gamma (efficientDet default gamma=1.5)
+hsv_h: 0.0  # image HSV-Hue augmentation (fraction)
+hsv_s: 0.0  # image HSV-Saturation augmentation (fraction)
+hsv_v: 0.0  # image HSV-Value augmentation (fraction)
+degrees: 0 # image rotation (+/- deg)
+translate: 0  # image translation (+/- fraction)
+scale: 0  # image scale (+/- gain)
+shear: 0.0  # image shear (+/- deg)
+perspective: 0.0  # image perspective (+/- fraction), range 0-0.001
+flipud: 0.3  # image flip up-down (probability)
+fliplr: 0.5  # image flip left-right (probability)
+mosaic: 0.5  # image mosaic (probability)
+mixup: 0.15  # image mixup (probability)
+copy_paste: 0.0  # image copy paste (probability)
+paste_in: 0.0  # image copy paste (probability), use 0 for faster training : cutout
+loss_ota: 0 #1 # use ComputeLossOTA, use 0 for faster training
+inversion: 0.5 #opposite temperature
+tir_channel_expansion: 0.3 #[0, 0.2, 0.5]
+img_percentile_removal: 0.3
+beta : 0.3
+random_perspective : 0

data/hyp.tir_od.tiny_aug_no_inversion.yaml

@@ -0,0 +1,36 @@
+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: 3.0  # warmup epochs (fractions ok)
+warmup_momentum: 0.8  # warmup initial momentum
+warmup_bias_lr: 0.001 #0.001  # warmup initial bias lr
+box: 0.05  # box loss gain
+cls: 0.5  # cls loss gain
+cls_pw: 1.0  # cls BCELoss positive_weight
+obj: 1.0  # obj loss gain (scale with pixels)
+obj_pw: 1.0  # obj BCELoss positive_weight
+iou_t: 0.20  # IoU training threshold
+anchor_t: 4.0  # anchor-multiple threshold
+anchors: 2  # anchors per output layer (0 to ignore)  @@HK was 3
+fl_gamma: 1.5 #1.5  # focal loss gamma (efficientDet default gamma=1.5)
+hsv_h: 0.0  # image HSV-Hue augmentation (fraction)
+hsv_s: 0.0  # image HSV-Saturation augmentation (fraction)
+hsv_v: 0.0  # image HSV-Value augmentation (fraction)
+degrees: 0 # image rotation (+/- deg)
+translate: 0  # image translation (+/- fraction)
+scale: 0  # image scale (+/- gain)
+shear: 0.0  # image shear (+/- deg)
+perspective: 0.0  # image perspective (+/- fraction), range 0-0.001
+flipud: 0.3  # image flip up-down (probability)
+fliplr: 0.5  # image flip left-right (probability)
+mosaic: 0.5  # image mosaic (probability)
+mixup: 0.15  # image mixup (probability)
+copy_paste: 0.0  # image copy paste (probability)
+paste_in: 0.0  # image copy paste (probability), use 0 for faster training : cutout
+loss_ota: 0 #1 # use ComputeLossOTA, use 0 for faster training
+inversion: 0 #opposite temperature
+tir_channel_expansion: 0.3 #[0, 0.2, 0.5]
+img_percentile_removal: 0.3
+beta : 0.3
+random_perspective : 0

data/hyp.tir_od.tiny_aug_no_inversion_paste_in.yaml

@@ -0,0 +1,36 @@
+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: 3.0  # warmup epochs (fractions ok)
+warmup_momentum: 0.8  # warmup initial momentum
+warmup_bias_lr: 0.001 #0.001  # warmup initial bias lr
+box: 0.05  # box loss gain
+cls: 0.5  # cls loss gain
+cls_pw: 1.0  # cls BCELoss positive_weight
+obj: 1.0  # obj loss gain (scale with pixels)
+obj_pw: 1.0  # obj BCELoss positive_weight
+iou_t: 0.20  # IoU training threshold
+anchor_t: 4.0  # anchor-multiple threshold
+anchors: 2  # anchors per output layer (0 to ignore)  @@HK was 3
+fl_gamma: 1.5 #1.5  # focal loss gamma (efficientDet default gamma=1.5)
+hsv_h: 0.0  # image HSV-Hue augmentation (fraction)
+hsv_s: 0.0  # image HSV-Saturation augmentation (fraction)
+hsv_v: 0.0  # image HSV-Value augmentation (fraction)
+degrees: 0 # image rotation (+/- deg)
+translate: 0  # image translation (+/- fraction)
+scale: 0  # image scale (+/- gain)
+shear: 0.0  # image shear (+/- deg)
+perspective: 0.0  # image perspective (+/- fraction), range 0-0.001
+flipud: 0.3  # image flip up-down (probability)
+fliplr: 0.5  # image flip left-right (probability)
+mosaic: 0.5  # image mosaic (probability)
+mixup: 0.15  # image mixup (probability)
+copy_paste: 0.0  # image copy paste (probability)
+paste_in: 0.3  # image copy paste (probability), use 0 for faster training : cutout
+loss_ota: 0 #1 # use ComputeLossOTA, use 0 for faster training
+inversion: 0 #opposite temperature
+tir_channel_expansion: 0.3 #[0, 0.2, 0.5]
+img_percentile_removal: 0.3
+beta : 0.3
+random_perspective : 0

data/hyp.tir_od.tiny_lr0_0p005.yaml

@@ -0,0 +1,36 @@
+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: 3.0  # warmup epochs (fractions ok)
+warmup_momentum: 0.8  # warmup initial momentum
+warmup_bias_lr: 0.001 #0.001  # warmup initial bias lr
+box: 0.05  # box loss gain
+cls: 0.5  # cls loss gain
+cls_pw: 1.0  # cls BCELoss positive_weight
+obj: 1.0  # obj loss gain (scale with pixels)
+obj_pw: 1.0  # obj BCELoss positive_weight
+iou_t: 0.20  # IoU training threshold
+anchor_t: 4.0  # anchor-multiple threshold
+anchors: 2  # anchors per output layer (0 to ignore)  @@HK was 3
+fl_gamma: 1.5 #1.5  # focal loss gamma (efficientDet default gamma=1.5)
+hsv_h: 0.0  # image HSV-Hue augmentation (fraction)
+hsv_s: 0.0  # image HSV-Saturation augmentation (fraction)
+hsv_v: 0.0  # image HSV-Value augmentation (fraction)
+degrees: 0.0  # image rotation (+/- deg)
+translate: 0.0  # image translation (+/- fraction)
+scale: 0.0  # image scale (+/- gain)
+shear: 0.0  # image shear (+/- deg)
+perspective: 0.0  # image perspective (+/- fraction), range 0-0.001
+flipud: 0.3  # image flip up-down (probability)
+fliplr: 0.5  # image flip left-right (probability)
+mosaic: 0.0  # image mosaic (probability)
+mixup: 0.0  # image mixup (probability)
+copy_paste: 0.0  # image copy paste (probability)
+paste_in: 0.0  # image copy paste (probability), use 0 for faster training : cutout
+loss_ota: 0 #1 # use ComputeLossOTA, use 0 for faster training
+inversion: 0.5 #opposite temperature
+drc_per_ch_percentile: [0, 0.2, 0.5]
+img_percentile_removal: 0.3
+beta : 0.3
+random_perspective : 0

data/hyp.tir_od.tiny_lr0_0p01.yaml

@@ -0,0 +1,36 @@
+lr0: 0.01 #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: 3.0  # warmup epochs (fractions ok)
+warmup_momentum: 0.8  # warmup initial momentum
+warmup_bias_lr: 0.001 #0.001  # warmup initial bias lr
+box: 0.05  # box loss gain
+cls: 0.5  # cls loss gain
+cls_pw: 1.0  # cls BCELoss positive_weight
+obj: 1.0  # obj loss gain (scale with pixels)
+obj_pw: 1.0  # obj BCELoss positive_weight
+iou_t: 0.20  # IoU training threshold
+anchor_t: 4.0  # anchor-multiple threshold
+anchors: 2  # anchors per output layer (0 to ignore)  @@HK was 3
+fl_gamma: 1.5 #1.5  # focal loss gamma (efficientDet default gamma=1.5)
+hsv_h: 0.0  # image HSV-Hue augmentation (fraction)
+hsv_s: 0.0  # image HSV-Saturation augmentation (fraction)
+hsv_v: 0.0  # image HSV-Value augmentation (fraction)
+degrees: 0.0  # image rotation (+/- deg)
+translate: 0.0  # image translation (+/- fraction)
+scale: 0.0  # image scale (+/- gain)
+shear: 0.0  # image shear (+/- deg)
+perspective: 0.0  # image perspective (+/- fraction), range 0-0.001
+flipud: 0.3  # image flip up-down (probability)
+fliplr: 0.5  # image flip left-right (probability)
+mosaic: 0.0  # image mosaic (probability)
+mixup: 0.0  # image mixup (probability)
+copy_paste: 0.0  # image copy paste (probability)
+paste_in: 0.0  # image copy paste (probability), use 0 for faster training : cutout
+loss_ota: 0 #1 # use ComputeLossOTA, use 0 for faster training
+inversion: 0.5 #opposite temperature
+drc_per_ch_percentile: [0, 0.2, 0.5]
+img_percentile_removal: 0.3
+beta : 0.3
+random_perspective : 0

data/hyp.tir_od.tiny_lr0_0p05.yaml

@@ -0,0 +1,36 @@
+lr0: 0.05 #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: 3.0  # warmup epochs (fractions ok)
+warmup_momentum: 0.8  # warmup initial momentum
+warmup_bias_lr: 0.001 #0.001  # warmup initial bias lr
+box: 0.05  # box loss gain
+cls: 0.5  # cls loss gain
+cls_pw: 1.0  # cls BCELoss positive_weight
+obj: 1.0  # obj loss gain (scale with pixels)
+obj_pw: 1.0  # obj BCELoss positive_weight
+iou_t: 0.20  # IoU training threshold
+anchor_t: 4.0  # anchor-multiple threshold
+anchors: 2  # anchors per output layer (0 to ignore)  @@HK was 3
+fl_gamma: 1.5 #1.5  # focal loss gamma (efficientDet default gamma=1.5)
+hsv_h: 0.0  # image HSV-Hue augmentation (fraction)
+hsv_s: 0.0  # image HSV-Saturation augmentation (fraction)
+hsv_v: 0.0  # image HSV-Value augmentation (fraction)
+degrees: 0.0  # image rotation (+/- deg)
+translate: 0.0  # image translation (+/- fraction)
+scale: 0.0  # image scale (+/- gain)
+shear: 0.0  # image shear (+/- deg)
+perspective: 0.0  # image perspective (+/- fraction), range 0-0.001
+flipud: 0.3  # image flip up-down (probability)
+fliplr: 0.5  # image flip left-right (probability)
+mosaic: 0.0  # image mosaic (probability)
+mixup: 0.0  # image mixup (probability)
+copy_paste: 0.0  # image copy paste (probability)
+paste_in: 0.0  # image copy paste (probability), use 0 for faster training : cutout
+loss_ota: 0 #1 # use ComputeLossOTA, use 0 for faster training
+inversion: 0.5 #opposite temperature
+drc_per_ch_percentile: [0, 0.2, 0.5]
+img_percentile_removal: 0.3
+beta : 0.3
+random_perspective : 0

data/tir_od.yaml

@@ -0,0 +1,17 @@
+# 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.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/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']

data/tir_od_test_set.yaml

@@ -0,0 +1,16 @@
+# COCO 2017 dataset http://cocodataset.org
+
+# download command/URL (optional)
+
+path: /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/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']