Yuxian Meng

@minstar `evaluate.py` actually run `self.validation_step` defined in `trainer.py`. I recommend to read pytorch-lightning documents to learn more about its mechanism and implement your own metrics

@phellonchen Hi, we found the current cnn/rcnn features for valid/test set is not correct, we will update it soon!

@phellonchen We updated cnn/rcnn feature of valid/test dataset today! We achieved 0.95/1.14/1.18 BLEU4 for NV/CV/FV, respectively. Other stats are coming soon.

@wjczf123 Yes, you're right. Only data in preprocessed_dir are used during training and inference.

@xiang-xiang-zhu Hi, I guess you are not familiar with [mmap format](https://numpy.org/doc/stable/reference/generated/numpy.memmap.html). We choose to use `mmap` instead of `npy` because `.npy` will load `np.array` to memory, but our feature file...

@xiang-xiang-zhu Yes, please refer to our [corresponding code](https://github.com/ShannonAI/OpenViDial/blob/main/video_dialogue_model/data/feature_dataset.py#L29).

@xiang-xiang-zhu Yes, both of your comments are right.

Did you try r=2 on MNIST data? I think this issue may caused by your toy batch. For example, if all your inputs in this toy batch are same while...

@JianboGuo Hi, I've run my code again. It seems working quite well. ```python train.py -batch_size=8 -lr=2e-2 -num_epochs=5 -r=2 -print_freq=5``` returns something like ``` batch:5, loss:0.3607, acc:0/8 batch:10, loss:0.3606, acc:4/8 batch:15,...

@JianboGuo Thanks for your feedback. It turns out that I ran the wrong code myself, I'm very sorry for that. I've reproduced this problem and trying to fix it. If...