Visualization code

[hq0709]

Thanks for your great work!

regarding this figure, could you please release the code of visualization?

Thanks!

[QtacierP]

Thanks for your attention. In fact, this visualization is based on the text to image cross-attention map in

PRIOR/codes/prior/models/prior.py

Line 179 in 086bf2e

text_to_local_image_embed, text_to_local_image_atten, image_to_local_text_embed, image_to_local_text_atten = self.local_cross_attention(local_image_embed, local_text_embed)

I will try my best to clean and release the related codes ASAP.

[hq0709]

Thanks! Your visualization code is very important to my current work. I appreciate it if you can release that!

[QtacierP]

Thank you. The existing visualization codes are quite messy and intertwined within the training pipeline. My plan is to create a demo.py file that can accept any text and image as input for zero-shot classification (including visualization). I will make the necessary updates once I complete my current work, which is an extended version of PRIOR.

[hq0709]

Could you open source the visualization code?

Many thanks!

[violet-42]

Thank you for your work. Could you make your visualization code public?