YE Zhou

Why choose "SoftmaxWithLoss" for "loss_attr"

4

Hi @peteanderson80 , Thanks for your sharing. There are some objects has multiple attributes, why choose "SoftmaxWithLoss" for "loss_attr"?

About data augmentation

2

Hi@LBH1024, it seems that the code about data augmentation is missing, which is essential to get better performance, can you share this code? Thanks!

About requirements

2

Hi@LBH1024, can you tell me the requirements~(such as the python and pytorch version) to run the code. Thanks!

In generator , the return value is (tf.tanh(h4)/2. + 0.5), which is fake image. The range of pixel value is [0,1] for fake image. However, the range of pixel value...

skipthought/coco_cap_uni_gru_word2vec.t7

1

Hi, @JonghwanMun , thanks for your excellent code. I have a little question to ask. Specifically, Did you obtain the skipthought/coco_cap_uni_gru_word2vec.t7 file by training the skipthought model? Can I just...

VGG-FCN t7 file

6

You had released the Res-101.t7 file to extract image features, can you release the VGG-FCN t7 file ? Thanks.

Thanks guys for your great great work. I have a question about the design of CustomCheckpointIO. When we want to save multiple model weights (including the last), we set [save_top_k=3](https://github.com/clovaai/donut/blob/217cffb111a57ebce1025ff84a722a8d9914e05b/train.py#L109)...

It seems that load_dataset is very slow to load about 11M images, How did you solve it？

1

Thanks for your great work@josianem @gwkrsrch. It seems that huggingface's load_dataset is very slow to load about 11M images, How did you solve it？