search

hitachi-rd-cv / qpic

Repo for CVPR2021 paper "QPIC: Query-Based Pairwise Human-Object Interaction Detection with Image-Wide Contextual Information"
Apache License 2.0
131 stars 33 forks source link

Why the pretrained VCOCO model has 81 object classes? #6

Closed weiyunfei closed 3 years ago

weiyunfei commented 3 years ago

When I tried to evaluate the VCOCO model you provide, I have to set the parameter --num_obj_classes to 81. What is the reason for this setting? And should I set --num_obj_classes 81 during training?

Thanks for your reply!

weiyunfei commented 3 years ago

I've checked the code, I found that the VCOCO model you released has an 82-way object head. I just wanner verify whether it is a technical error.

tamtamz commented 3 years ago

Thank you for pointing it out. The correct number is 81. The number of object classes in the training command for V-COCO, which was 80, was incorrect, and I fixed it to 81. The one additional class is for the missing object case in the V-COCO dataset, where a person is acting, but a target object is missing.

weiyunfei commented 3 years ago

Thanks!