Closed fusang1337 closed 2 years ago
(next mAP calculation at 6175 iterations) Last accuracy mAP@0.50 = 62.27 %, best = 66.48 % 6000: 1.022357, 1.471952 avg loss, 0.000010 rate, 4.679000 seconds, 192000 images, 0.091826 hours left Resizing to initial size: 416 x 416 try to allocate additional workspace_size = 52.43 MB CUDA allocate done!
calculation mAP (mean average precision)... Detection layer: 139 - type = 28 Detection layer: 150 - type = 28 Detection layer: 161 - type = 28 260 detections_count = 1069, unique_truth_count = 413 class_id = 0, name = fire, ap = 61.61% (TP = 258, FP = 106)
for conf_thresh = 0.25, precision = 0.71, recall = 0.62, F1-score = 0.66 for conf_thresh = 0.25, TP = 258, FP = 106, FN = 155, average IoU = 54.18 %
IoU threshold = 50 %, used Area-Under-Curve for each unique Recall mean average precision (mAP@0.50) = 0.616056, or 61.61 % Total Detection Time: 9 Seconds
感觉指标是不是有点低啊=-=
看实际效果,不惟精度论
实际测试了几个视频,主观上效果都挺不错的。但是不知道怎么量化指标,我想批量测试照片,然后得到指标,批量测试照片不知道怎么做=-=
(next mAP calculation at 6175 iterations) Last accuracy mAP@0.50 = 62.27 %, best = 66.48 % 6000: 1.022357, 1.471952 avg loss, 0.000010 rate, 4.679000 seconds, 192000 images, 0.091826 hours left Resizing to initial size: 416 x 416 try to allocate additional workspace_size = 52.43 MB CUDA allocate done!
calculation mAP (mean average precision)... Detection layer: 139 - type = 28 Detection layer: 150 - type = 28 Detection layer: 161 - type = 28 260 detections_count = 1069, unique_truth_count = 413 class_id = 0, name = fire, ap = 61.61% (TP = 258, FP = 106)
for conf_thresh = 0.25, precision = 0.71, recall = 0.62, F1-score = 0.66 for conf_thresh = 0.25, TP = 258, FP = 106, FN = 155, average IoU = 54.18 %
IoU threshold = 50 %, used Area-Under-Curve for each unique Recall mean average precision (mAP@0.50) = 0.616056, or 61.61 % Total Detection Time: 9 Seconds
感觉指标是不是有点低啊=-=![chart_yolov4-fire](https://user-images.githubusercontent.com/17375049/128688262-317c8356-9fc6-41ed-9ea7-98920751d473.png)