目录
//8月16 v7tiny byd 10k 512X512
Tensor Cores are used.
Last accuracy mAP@0.50 = 90.72 %, best = 90.81 %
10000: 3.239424, 3.549506 avg loss, 0.000026 rate, 1.057359 seconds, 640000 images, 0.031961 hours left
calculation mAP (mean average precision)...
Detection layer: 90 - type = 28
Detection layer: 94 - type = 28
Detection layer: 98 - type = 28
360
detections_count = 23087, unique_truth_count = 2448
class_id = 0, name = Person, ap = 86.57% (TP = 299, FP = 58)
class_id = 1, name = AythyaBaeri, ap = 96.70% (TP = 854, FP = 85)
class_id = 2, name = Osprey, ap = 84.21% (TP = 457, FP = 106)
class_id = 3, name = Lotus, ap = 90.14% (TP = 305, FP = 37)
class_id = 4, name = Boat, ap = 95.73% (TP = 173, FP = 31)
for conf_thresh = 0.25, precision = 0.87, recall = 0.85, F1-score = 0.86
for conf_thresh = 0.25, TP = 2088, FP = 317, FN = 360, average IoU = 67.59 %
IoU threshold = 50 %, used Area-Under-Curve for each unique Recall
mean average precision (mAP@0.50) = 0.906681, or 90.67 %
Total Detection Time: 17 Seconds
Set -points flag:
`-points 101` for MS COCO
`-points 11` for PascalVOC 2007 (uncomment `difficult` in voc.data)
`-points 0` (AUC) for ImageNet, PascalVOC 2010-2012, your custom dataset
mean_average_precision (mAP@0.50) = 0.906681
Saving weights to /home/jykj/wjy/01_yolov4/backup//yolov7-tiny_byd_10k_10000.weights
Saving weights to /home/jykj/wjy/01_yolov4/backup//yolov7-tiny_byd_10k_last.weights
Saving weights to /home/jykj/wjy/01_yolov4/backup//yolov7-tiny_byd_10k_final.weights
If you want to train from the beginning, then use flag in the end of training command: -clear
//8.16 yolov7-tiny_byd_20k
Tensor Cores are used.
Last accuracy mAP@0.50 = 91.45 %, best = 91.54 %
20000: 3.006380, 3.025432 avg loss, 0.000026 rate, 1.758795 seconds, 1280000 images, 0.049424 hours left
calculation mAP (mean average precision)...
Detection layer: 90 - type = 28
Detection layer: 94 - type = 28
Detection layer: 98 - type = 28
360
detections_count = 18772, unique_truth_count = 2448
class_id = 0, name = Person, ap = 88.25% (TP = 301, FP = 50)
class_id = 1, name = AythyaBaeri, ap = 96.84% (TP = 862, FP = 80)
class_id = 2, name = Osprey, ap = 84.80% (TP = 468, FP = 107)
class_id = 3, name = Lotus, ap = 90.51% (TP = 333, FP = 40)
class_id = 4, name = Boat, ap = 96.62% (TP = 176, FP = 26)
for conf_thresh = 0.25, precision = 0.88, recall = 0.87, F1-score = 0.88
for conf_thresh = 0.25, TP = 2140, FP = 303, FN = 308, average IoU = 68.07 %
IoU threshold = 50 %, used Area-Under-Curve for each unique Recall
mean average precision (mAP@0.50) = 0.914050, or 91.41 %
Total Detection Time: 19 Seconds
Set -points flag:
`-points 101` for MS COCO
`-points 11` for PascalVOC 2007 (uncomment `difficult` in voc.data)
`-points 0` (AUC) for ImageNet, PascalVOC 2010-2012, your custom dataset
mean_average_precision (mAP@0.50) = 0.914050
Saving weights to /home/jykj/wjy/01_yolov4/backup//yolov7-tiny_byd_20k_20000.weights
Saving weights to /home/jykj/wjy/01_yolov4/backup//yolov7-tiny_byd_20k_last.weights
Saving weights to /home/jykj/wjy/01_yolov4/backup//yolov7-tiny_byd_20k_final.weights
If you want to train from the beginning, then use flag in the end of training command: -clear
//8月16日 yolov4-tiny_byd_10k
Tensor Cores are used.
Last accuracy mAP@0.50 = 82.73 %, best = 82.90 %
10000: 0.366318, 0.636723 avg loss, 0.000026 rate, 0.666808 seconds, 640000 images, 0.028370 hours left
calculation mAP (mean average precision)...
Detection layer: 30 - type = 28
Detection layer: 37 - type = 28
360
detections_count = 4854, unique_truth_count = 2448
class_id = 0, name = Person, ap = 77.41% (TP = 271, FP = 56)
class_id = 1, name = AythyaBaeri, ap = 96.79% (TP = 836, FP = 89)
class_id = 2, name = Osprey, ap = 84.41% (TP = 472, FP = 135)
class_id = 3, name = Lotus, ap = 62.41% (TP = 195, FP = 15)
class_id = 4, name = Boat, ap = 92.47% (TP = 164, FP = 28)
for conf_thresh = 0.25, precision = 0.86, recall = 0.79, F1-score = 0.82
for conf_thresh = 0.25, TP = 1938, FP = 323, FN = 510, average IoU = 66.15 %
IoU threshold = 50 %, used Area-Under-Curve for each unique Recall
mean average precision (mAP@0.50) = 0.826958, or 82.70 %
Total Detection Time: 15 Seconds
Set -points flag:
`-points 101` for MS COCO
`-points 11` for PascalVOC 2007 (uncomment `difficult` in voc.data)
`-points 0` (AUC) for ImageNet, PascalVOC 2010-2012, your custom dataset
mean_average_precision (mAP@0.50) = 0.826958
Saving weights to /home/jykj/wjy/01_yolov4/backup//yolov4-tiny_byd_10k_10000.weights
Saving weights to /home/jykj/wjy/01_yolov4/backup//yolov4-tiny_byd_10k_last.weights
Saving weights to /home/jykj/wjy/01_yolov4/backup//yolov4-tiny_byd_10k_final.weights
If you want to train from the beginning, then use flag in the end of training command: -clear
//yolof_byd10k
(next mAP calculation at 10030 iterations)
Last accuracy mAP@0.5 = 84.63 %, best = 85.01 %
10000: 1.414221, 1.350827 avg loss, 0.000010 rate, 1.637153 seconds, 640000 images, 0.052448 hours left
calculation mAP (mean average precision)...
Detection layer: 121 - type = 28
Detection layer: 130 - type = 28
360
detections_count = 9504, unique_truth_count = 2448
class_id = 0, name = Person, ap = 77.20% (TP = 302, FP = 186)
class_id = 1, name = AythyaBaeri, ap = 97.10% (TP = 861, FP = 173)
class_id = 2, name = Osprey, ap = 79.58% (TP = 476, FP = 307)
class_id = 3, name = Lotus, ap = 78.51% (TP = 342, FP = 202)
class_id = 4, name = Boat, ap = 92.55% (TP = 170, FP = 74)
for conf_thresh = 0.25, precision = 0.70, recall = 0.88, F1-score = 0.78
for conf_thresh = 0.25, TP = 2151, FP = 942, FN = 297, average IoU = 51.75 %
IoU threshold = 50 %, used Area-Under-Curve for each unique Recall
mean average precision (mAP@0.50) = 0.849861, or 84.99 %
Total Detection Time: 25 Seconds
Set -points flag:
`-points 101` for MS COCO
`-points 11` for PascalVOC 2007 (uncomment `difficult` in voc.data)
`-points 0` (AUC) for ImageNet, PascalVOC 2010-2012, your custom dataset
mean_average_precision (mAP@0.5) = 0.849861
Saving weights to /home/jykj/wjy/05_Yolo-Fastest-master/backup//yolof_byd10k_10000.weights
Saving weights to /home/jykj/wjy/05_Yolo-Fastest-master/backup//yolof_byd10k_last.weights
Saving weights to /home/jykj/wjy/05_Yolo-Fastest-master/backup//yolof_byd10k_final.weights
If you want to train from the beginning, then use flag in the end of training command: -clear
//yolov4-tiny_byd_20k
(next mAP calculation at 20095 iterations)
Tensor Cores are used.
Last accuracy mAP@0.50 = 83.16 %, best = 83.38 %
20000: 0.444506, 0.581204 avg loss, 0.000026 rate, 1.172310 seconds, 1280000 images, 0.035644 hours left
calculation mAP (mean average precision)...
Detection layer: 30 - type = 28
Detection layer: 37 - type = 28
360
detections_count = 4179, unique_truth_count = 2448
class_id = 0, name = Person, ap = 76.52% (TP = 273, FP = 53)
class_id = 1, name = AythyaBaeri, ap = 97.12% (TP = 844, FP = 87)
class_id = 2, name = Osprey, ap = 84.75% (TP = 485, FP = 125)
class_id = 3, name = Lotus, ap = 63.60% (TP = 196, FP = 13)
class_id = 4, name = Boat, ap = 93.71% (TP = 166, FP = 27)
for conf_thresh = 0.25, precision = 0.87, recall = 0.80, F1-score = 0.83
for conf_thresh = 0.25, TP = 1964, FP = 305, FN = 484, average IoU = 67.12 %
IoU threshold = 50 %, used Area-Under-Curve for each unique Recall
mean average precision (mAP@0.50) = 0.831374, or 83.14 %
Total Detection Time: 20 Seconds
Set -points flag:
`-points 101` for MS COCO
`-points 11` for PascalVOC 2007 (uncomment `difficult` in voc.data)
`-points 0` (AUC) for ImageNet, PascalVOC 2010-2012, your custom dataset
mean_average_precision (mAP@0.50) = 0.831374
Saving weights to /home/jykj/wjy/01_yolov4/backup//yolov4-tiny_byd_20k_20000.weights
Saving weights to /home/jykj/wjy/01_yolov4/backup//yolov4-tiny_byd_20k_last.weights
Saving weights to /home/jykj/wjy/01_yolov4/backup//yolov4-tiny_byd_20k_final.weights
//yolof_byd20k
(next mAP calculation at 20080 iterations)
Last accuracy mAP@0.5 = 87.61 %, best = 87.89 %
20000: 1.016090, 1.103532 avg loss, 0.000010 rate, 1.440473 seconds, 1280000 images, 0.044317 hours left
calculation mAP (mean average precision)...
Detection layer: 121 - type = 28
Detection layer: 130 - type = 28
360
detections_count = 7731, unique_truth_count = 2448
class_id = 0, name = Person, ap = 82.98% (TP = 318, FP = 169)
class_id = 1, name = AythyaBaeri, ap = 97.40% (TP = 858, FP = 131)
class_id = 2, name = Osprey, ap = 81.36% (TP = 480, FP = 269)
class_id = 3, name = Lotus, ap = 80.92% (TP = 345, FP = 160)
class_id = 4, name = Boat, ap = 95.14% (TP = 174, FP = 63)
for conf_thresh = 0.25, precision = 0.73, recall = 0.89, F1-score = 0.80
for conf_thresh = 0.25, TP = 2175, FP = 792, FN = 273, average IoU = 55.11 %
IoU threshold = 50 %, used Area-Under-Curve for each unique Recall
mean average precision (mAP@0.50) = 0.875596, or 87.56 %
Total Detection Time: 21 Seconds
Set -points flag:
`-points 101` for MS COCO
`-points 11` for PascalVOC 2007 (uncomment `difficult` in voc.data)
`-points 0` (AUC) for ImageNet, PascalVOC 2010-2012, your custom dataset
mean_average_precision (mAP@0.5) = 0.875596
Saving weights to /home/jykj/wjy/05_Yolo-Fastest-master/backup/yolof_byd20k_20000.weights
Saving weights to /home/jykj/wjy/05_Yolo-Fastest-master/backup/yolof_byd20k_last.weights
Saving weights to /home/jykj/wjy/05_Yolo-Fastest-master/backup/yolof_byd20k_final.weights
//yolov3-tiny_byd_10k
(next mAP calculation at 10045 iterations)
Tensor Cores are used.
Last accuracy mAP@0.50 = 78.47 %, best = 79.01 %
10000: 1.606564, 2.013261 avg loss, 0.000010 rate, 0.731317 seconds, 640000 images, 0.023691 hours left
Resizing to initial size: 512 x 512 try to allocate additional workspace_size = 84.87 MB
CUDA allocate done!
calculation mAP (mean average precision)...
Detection layer: 16 - type = 28
Detection layer: 23 - type = 28
360
detections_count = 7345, unique_truth_count = 2448
class_id = 0, name = Person, ap = 72.10% (TP = 268, FP = 86)
class_id = 1, name = AythyaBaeri, ap = 95.23% (TP = 833, FP = 115)
class_id = 2, name = Osprey, ap = 73.94% (TP = 397, FP = 129)
class_id = 3, name = Lotus, ap = 61.19% (TP = 288, FP = 129)
class_id = 4, name = Boat, ap = 91.88% (TP = 163, FP = 30)
for conf_thresh = 0.25, precision = 0.80, recall = 0.80, F1-score = 0.80
for conf_thresh = 0.25, TP = 1949, FP = 489, FN = 499, average IoU = 58.86 %
IoU threshold = 50 %, used Area-Under-Curve for each unique Recall
mean average precision (mAP@0.50) = 0.788675, or 78.87 %
Total Detection Time: 11 Seconds
Set -points flag:
`-points 101` for MS COCO
`-points 11` for PascalVOC 2007 (uncomment `difficult` in voc.data)
`-points 0` (AUC) for ImageNet, PascalVOC 2010-2012, your custom dataset
mean_average_precision (mAP@0.50) = 0.788675
Saving weights to /home/jykj/wjy/01_yolov4/backup/yolov3-tiny_byd_10k_10000.weights
Saving weights to /home/jykj/wjy/01_yolov4/backup/yolov3-tiny_byd_10k_last.weights
Saving weights to /home/jykj/wjy/01_yolov4/backup/yolov3-tiny_byd_10k_final.weights
//yolov3-tiny_byd_20k
(next mAP calculation at 20095 iterations)
Tensor Cores are used.
Last accuracy mAP@0.50 = 81.45 %, best = 81.59 %
20000: 1.467952, 1.765743 avg loss, 0.000010 rate, 0.855782 seconds, 1280000 images, 0.029438 hours left
Resizing to initial size: 512 x 512 try to allocate additional workspace_size = 103.81 MB
CUDA allocate done!
calculation mAP (mean average precision)...
Detection layer: 16 - type = 28
Detection layer: 23 - type = 28
360
detections_count = 6335, unique_truth_count = 2448
class_id = 0, name = Person, ap = 76.13% (TP = 290, FP = 94)
class_id = 1, name = AythyaBaeri, ap = 96.23% (TP = 850, FP = 101)
class_id = 2, name = Osprey, ap = 73.74% (TP = 407, FP = 139)
class_id = 3, name = Lotus, ap = 65.32% (TP = 315, FP = 138)
class_id = 4, name = Boat, ap = 94.50% (TP = 169, FP = 24)
for conf_thresh = 0.25, precision = 0.80, recall = 0.83, F1-score = 0.82
for conf_thresh = 0.25, TP = 2031, FP = 496, FN = 417, average IoU = 59.91 %
IoU threshold = 50 %, used Area-Under-Curve for each unique Recall
mean average precision (mAP@0.50) = 0.811851, or 81.19 %
Total Detection Time: 16 Seconds
Set -points flag:
`-points 101` for MS COCO
`-points 11` for PascalVOC 2007 (uncomment `difficult` in voc.data)
`-points 0` (AUC) for ImageNet, PascalVOC 2010-2012, your custom dataset
mean_average_precision (mAP@0.50) = 0.811851
Saving weights to /home/jykj/wjy/01_yolov4/backup/yolov3-tiny_byd_20k_20000.weights
Saving weights to /home/jykj/wjy/01_yolov4/backup/yolov3-tiny_byd_20k_last.weights
Saving weights to /home/jykj/wjy/01_yolov4/backup/yolov3-tiny_byd_20k_final.weights
If you want to train from the beginning, then use flag in the end of training command: -clear