search

Charles-Xie / CQL

Code for our paper "Category Query Learning for Human-Object Interaction Classification" (CVPR2023)
https://arxiv.org/abs/2303.14005
Apache License 2.0
36 stars 4 forks source link

GEN-VLKT+CQL #7

Open paihuai00 opened 1 year ago

paihuai00 commented 1 year ago

When applying the CQL method to GEN-VLKT, how should we handle the CLIP weight for verb classification?

Can you provide the relevant code?

Charles-Xie commented 1 year ago

When applying CQL to GEN-VLKT, we use CLIP weight to initialize the category queries. I will upload the code for this part after finishing a recent deadline :smiley:

paihuai00 commented 1 year ago

When applying CQL to GEN-VLKT, we use CLIP weight to initialize the category queries. I will upload the code for this part after finishing a recent deadline 😃

How's it going?

xiguahong commented 1 year ago

Thank you for your work. It has been three months. Can you release the code for GENVLKT+CQL now? I'm having trouble with its performance.

kkayla99 commented 1 year ago

@Charles-Xie Hello, I would like to inquire whether the CLIP weight used to initialize the category queries is derived from the HOI triplet text or just from interaction text, and how? What is the coefficient for the image loss in GEN? Could you update some code examples?