Yun Chen

Well, I'll do it if I have some extral time, but busy lately.

这个可能和环境的配置有关，建议先在单卡的环境下调试通过

抱歉，我暂时没法check， 因为没有测试集的label，你能试试 在https://github.com/chenyuntc/pytorch-book/blob/master/chapter6-%E5%AE%9E%E6%88%98%E6%8C%87%E5%8D%97/main.py#L26 增加model=model.eval()吗？

Actually I use another kind of early stop: - if val acc rises: save the model, keep training - if val acc drops: load previous best model, decay lr see...

It seems that something was wrong with your data preprocessing.

It could be done by implementing your own data set, see example of: https://github.com/chenyuntc/simple-faster-rcnn-pytorch/blob/master/data/voc_dataset.py it should return - img: numpy array in the shape of (3,H,W) - label: a int...

As indicated in the error message: ``` label.append(VOC_BBOX_LABEL_NAMES.index(name)) ValueError: tuple.index(x): x not in tuple ``` I think the annotation file is somewhat `wrong` so that the `name` is not in...

- `shep` comes from https://github.com/chenyuntc/simple-faster-rcnn-pytorch/blob/master/utils/vis_tool.py#L31 these are the brief names(i.e. sheep->shep cat->c dog->d) I being lazy... - As for the sample images doesn't change, I don't have ideas.

> what is the size (resolution) of images which I should use for training my network It depends. In my experiments I crop image in [`preprocess`](https://github.com/chenyuntc/simple-faster-rcnn-pytorch/blob/master/data/dataset.py#L40)

I think maybe resnet101 is difficult to train. [This](https://github.com/ruotianluo/pytorch-faster-rcnn/blob/master/README.md) maybe helpful. >For Resnets, we fix the first block (total 4) when fine-tuning the network, and only use crop_and_resize to resize...