training doesn't cost time about fcos

[tianzhi0549]

@1ewandovski It seems that the training ends very quickly. Can you check your dataset? I guess your dataset contains no images or none of the images is read successfully.

[1ewandovski]

@tianzhi0549 I've checked my dataset and didn't find any problem with it. Actually, during evaluation, "coco_2019_val" can be load correctly,which also used for training.
By the way, I can train on these datasets using original maskrcnn-benchmark

INFO: Start evaluation on coco_2019_val dataset(121 images).
100%|????????????????????????????????????????????????| 121/121 [00:04<00:00, 25.15it/s]
2019-05-05 21:45:29,280 maskrcnn_benchmark.inference INFO: Total run time: 0:00:04.816332 (0.039804397535718176 s / img per device, on 1 devices)
2019-05-05 21:45:29,281 maskrcnn_benchmark.inference INFO: Model inference time: 0:00:04.087672 (0.03378241121276351 s / img per device, on 1 devices)
2019-05-05 21:45:29,306 maskrcnn_benchmark.inference INFO: Preparing results for COCO format
2019-05-05 21:45:29,306 maskrcnn_benchmark.inference INFO: Preparing bbox results
2019-05-05 21:45:29,307 maskrcnn_benchmark.inference INFO: Evaluating predictions
Running per image evaluation...
Evaluate annotation type bbox
DONE (t=0.06s).
Accumulating evaluation results...
DONE (t=0.02s).

[tianzhi0549]

@1ewandovski Probably because pre_train/mymodel.pth contains the iteration number. Therefore, the training finishes immediately since the iteration is initialized by the iteration stored in the model. Please try to manually remove iteration in the model with torch.load() and torch.save() APIs.

[1ewandovski]

@tianzhi0549 thank you very much! i've solved the problem after remove iteration. i'll close the issue.