Open LIUTAOGE opened 1 year ago
You might want to adjust the learning rate according to the batch size you are using following https://github.com/facebookresearch/Detectron/blob/main/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14
您可能想根据您使用的批量大小调整学习率,如下https://github.com/facebookresearch/Detectron/blob/main/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml#L14
Hello, can you share the pretrain model? The original link has expired.
INFO:maskrcnn_benchmark.trainer:eta: 10:55:48 iter: 5100 loss: nan (nan) loss_obj_classifier: 0.0000 (0.0000) loss_pred_classifier: nan (nan) time: 1.1246 (1.1275) data: 0.0462 (0.0644) lr: 0.000467 max mem: 9431
SOLVER: BASE_LR: 0.001 WEIGHT_DECAY: 0.0001 MAX_ITER: 40000 STEPS: (50000,) IMS_PER_BATCH: 16 CHECKPOINT_PERIOD: 10000