yiqilin

Currently, we read images directly from the original ImageNet. To adapt to a format like 0153282900005.jpg, the code needs some modifications.

In my view, attention rpn requires high-quality (large and full objects are preferred) support instances for training and needs to carefully set the fine-tuning parameters, since attention rpn tends to...

If the classes are changed, the `ALL_CLASSES` needs to be modified too.

The config file seems right. I'm not sure what causes this problem since the fine-tuning only train the bbox classification head. We will look into it.

Thank you for your feedback. We will fix it as soon as possible.

This bug has been fixed in [here](https://github.com/open-mmlab/mmfewshot/pull/52).

FSCE paper and released code both mention that the CPE loss does not work well in coco. So we also do not provide the result on coco with CPE loss....