vanpersie32

yeah, it takes much time to preprocess all images

@gujiuxiang Thank you

I met the same problem

@dastan92 Hey, have you replicate the result in the paper?

@erastogi It seems that you should add require 'cudnn' in script ./000_data_construction/001_resNet101_conv_feat/extract_resnet_feat.lua.

I have implemented extracting feature in batch, but seems in vain. It seems that torch.save consumes much time. I don't think it a good idea to save the feature map...

@JonghwanMun The reason why feature extraction is so slow is because you collect garbage too often. Here Here Here [here](https://github.com/JonghwanMun/TextguidedATT/blob/master/000_data_construction/001_resNet101_conv_feat/extract_resnet_feat.lua#L111)

> hello, could you please make your code public

@phellonchen Hi，have you succeeded pretraining the model from scratch?

> This loss has a normal decrease as expected. Note that the logged loss is averaged within a short window, so some fluctuation is normal. Thanks for your reply. I...