mmengine - ERROR - /opt/conda/envs/mmyolo/lib/python3.8/site-packages/mmdet/evaluation/metrics/coco_metric.py - compute_metrics - 437 - The testing results of the whole dataset is empty.

01/24 14:31:08 - mmengine - INFO - Epoch(train)   [1][ 50/740]  lr: 2.2072e-04  eta: 1 day, 23:41:25  time: 0.4641  data_time: 0.2087  memory: 11131  grad_norm: 6540.1456  loss: 683.7374  loss_cls: 683.7374  loss_bbox: 0.0000  loss_dfl: 0.0000
01/24 14:31:27 - mmengine - INFO - Epoch(train)   [1][100/740]  lr: 4.4595e-04  eta: 1 day, 19:31:41  time: 0.3832  data_time: 0.1878  memory: 5650  grad_norm: 19.7268  loss: 5.3739  loss_cls: 5.3739  loss_bbox: 0.0000  loss_dfl: 0.0000
...
01/24 14:39:14 - mmengine - INFO - Epoch(train)   [2][550/740]  lr: 5.7948e-03  eta: 1 day, 16:28:22  time: 0.3727  data_time: 0.1775  memory: 5650  grad_norm: 0.0272  loss: 0.0001  loss_cls: 0.0001  loss_bbox: 0.0000  loss_dfl: 0.0000
01/24 14:39:33 - mmengine - INFO - Epoch(train)   [2][600/740]  lr: 6.0196e-03  eta: 1 day, 16:24:20  time: 0.3790  data_time: 0.1833  memory: 5650  grad_norm: 0.0253  loss: 0.0001  loss_cls: 0.0001  loss_bbox: 0.0000  loss_dfl: 0.0000
01/24 14:39:54 - mmengine - INFO - Epoch(train)   [2][650/740]  lr: 6.2444e-03  eta: 1 day, 16:26:04  time: 0.4039  data_time: 0.2029  memory: 5650  grad_norm: 0.0245  loss: 0.0000  loss_cls: 0.0000  loss_bbox: 0.0000  loss_dfl: 0.0000

MMYOLO에서 yolov8을 사용하기 위해 MMYOLO v0.4.0으로 업데이트 후 학습을 돌려보니 loss_bbox와 loss_dfl(distribution focal loss)가 0으로 나오고 2epoch만에 모든 loss가 0이 나오게 됩니다.

01/24 14:55:41 - mmengine - INFO - Evaluating bbox...
Loading and preparing results...
01/24 14:55:41 - mmengine - ERROR - /opt/conda/envs/mmyolo/lib/python3.8/site-packages/mmdet/evaluation/metrics/coco_metric.py - compute_metrics - 437 - The testing results of the whole dataset is empty.
01/24 14:55:41 - mmengine - INFO - Epoch(val) [5][185/185]  
01/24 14:55:41 - mmengine - WARNING - Since `metrics` is an empty dict, the behavior to save the best checkpoint will be skipped in this evaluation.

또 evaluation metric이 bbox인데 loss_bbox가 계산이 되지 않아 위와 같은 에러가 발생합니다.

boostcampaitech4lv23cv3 / final-project-level3-cv-17

mmengine - ERROR - /opt/conda/envs/mmyolo/lib/python3.8/site-packages/mmdet/evaluation/metrics/coco_metric.py - compute_metrics - 437 - The testing results of the whole dataset is empty. #26