lupantech / iconqa Goto Github PK

Dear authors,
I have some trouble reproducing the result in the paper using the training code.
I followed the instructions in this repository and trained 3 models for 3 IconQA subtasks. I used the default training arguments. However, I am not able to reproduce the result in the paper.

My reproduce result:

• choose_img: 79.038
• choose_txt: 67.369
• fill_in_blank: 79.467

The result from the paper:

• choose_img: 82.66
• choose_txt: 75.19
• fill_in_blank: 83.62

Could you provide the training arguments that I could use to reproduce the result in the paper?

Thank you very much.
Best regards.

I would like to use your pre-trained Resnet model on the Icon645 dataset for classification.

Using your pre-trained Icon645 Resnet model, I was able to output the 377-dimension softmax vector for classification (for 377 classes).

I would like to know the mapping from the class ID to the class name in your Icon645 Resnet model.
For example:
1 --> apple
2 --> acorn
...

Is this mapping available?
Thank you very much.