Questions about the Image Captioning Code

[AlonzoLeeeooo]

Hi @pinakinathc ,

Thanks for the implementation of the excellent work. I have a few questions about your code implementations. It would be really kind of you if you could answer them:

(1) As is mentioned in the paper, you said that you fine-tune LNFMM (the image captioning model) upon FS-COCO dataset. But the code image-captioning/datasets.py in class OursScene() does not return capslen item. I look up the whole codebase and find that it is required by the decoder part of LNFMM. Does it mean that you only fine-tune the encoder part of LNFMM model?

(2) I checked the GitHub repository of LNFMM and downloaded their pre-trained model weights. I plan to load the weights with your code but it ran into an error RuntimeError: unexpected EOF, expected 170260432 more bytes. The file might be corrupted. Do you re-train the image captioning model? Do you meet any kind of problem like this during your experiments?

Regards,
Chang

[pinakinathc]

hey Chang
when i used the LNFMM code, their trained model was not available. I basically used their code and retrained myself (albeit with some minor modifications). I can go through the LNFMM module once more if this issue persists.

[AlonzoLeeeooo]

Hi @pinakinathc ,

Thanks for replying! Look forwards to your further progress. Please feel free to @ me whenever you need disscussions. I would try to reply.

Regards,
Chang