evaluation during training

[seacloud-0420]

Does this code support evaluation during training? If I want to evaluate while training, what should I do?

[csuhan]

Yes.

• change the following settings to val set https://github.com/csuhan/s2anet/blob/d2870963815b10c8f7796577e30b7c3b14e47ac6/configs/dota/s2anet_r50_fpn_1x_dota.py#L133-L135

• add --validate to your training command https://github.com/csuhan/s2anet/blob/d2870963815b10c8f7796577e30b7c3b14e47ac6/tools/train.py#L27

[seacloud-0420]

Thank you for your answer, I have changed it as you said, but the training appears: NoteImplementedError:Built-in validation is not implemented yet in not_distributed training. Use distributed training or test.py and *eval.py scripts instead. Also, I made my dataset according to the format of the DOTA dataset, the object in my dataset is relatively large and it is not suitable for split, I did not split my dataset. I got an error when I tested my own dataset, as shown in the image below. Is this code not suitable for the uncut DOTA data format, if so how do I need to change it?

[csuhan]

Ohh I'm sorry. I checked again, and this codebase does not support eval during training. It just support testing-then-evaluation (in one command) if you setup the evaluation config.

If the images in your dataset is small and do not need to split, you can comment L26-27 https://github.com/csuhan/s2anet/blob/d2870963815b10c8f7796577e30b7c3b14e47ac6/mmdet/datasets/dota.py#L20-L34

Besides, you many need to change something in your result files, see this issue