A little confusion about training

[lenka844]

Thanks for released the brilliant work again! When i was in traing process, i found that if i resume the training, the process will rsume from the saved checkpoint, but the self.use_box will return to the False, i mean the type of crop will becam to random crop, i was confused whether that will bother the end of training? or should i set the warmup_epochs to zero, when i resume trainging, in order to make sure that the type of crop is same as previously? Thanks for your reply.

[xyupeng]

Hi, train_set.use_box is determined here https://github.com/xyupeng/ContrastiveCrop/blob/6bd54c433dd05e817cd3d3adcd7d61b4bb12c811/DDP_moco_ccrop.py#L267

The checkpoint will save the epoch and when it resumes the mode will continue as CCrop if the epoch is larger than warmup_epochs.

[lenka844]

Hi, train_set.use_box is determined here

https://github.com/xyupeng/ContrastiveCrop/blob/6bd54c433dd05e817cd3d3adcd7d61b4bb12c811/DDP_moco_ccrop.py#L267

The checkpoint will save the epoch and when it resumes the mode will continue as CCrop if the epoch is larger than warmup_epochs.

Thanks for replying. I 've noticed this part of code. But I thought this code means that no matter how epochs I start to resume, it must through the warmup epochs to make sure the epoch >= cfg.warmup_epochs + start_epoch, and that also means i have to use random crop instead of Ccrop in the checkpoints. let me give u a example: when i beakdown at 150eochs, and my warmup epochs set to 100 epochs, it already using the Ccrop, but when i resume the training, the start_epoch in the code became to 150, and the 'epoch 'in the formula——epoch >= cfg.warmup_epochs + start_epoch， also became to 150 epoch, so the condition to use the Ccrop has yet to be met, it has to use the Random crop warmup epochs. And i try to verify this situation, it really exists，when i resume, it can't continue use the Ccrop. I wondering if it is my operation problems。

[xyupeng]

I think you are correct. I have fixed it like this: https://github.com/xyupeng/ContrastiveCrop/blob/305ff6d06f7e81a5a9f7f80480d81adc11436f9b/DDP_moco_ccrop.py#L267 I think now the logic is correct. Thank you very much for finding this problem :)

[lenka844]

I think you are correct. I have fixed it like this:

https://github.com/xyupeng/ContrastiveCrop/blob/305ff6d06f7e81a5a9f7f80480d81adc11436f9b/DDP_moco_ccrop.py#L267

I think now the logic is correct. Thank you very much for finding this problem :)

Thanks for replying again！