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

number of decoder layers impact #10

Closed jihwanp closed 3 years ago

jihwanp commented 3 years ago

Hi, thanks for interesting work. Does the number of decoder layers have a significant impact on performance like in detr? And I cant find how much v100 you've used. maybe 8?

tamtamz commented 3 years ago

Thanks for your comments. We did not investigate the impact of the number of decoder layers. We used 8 GPUs.

jihwanp commented 3 years ago

@tamtamz thanks i have one more question. what is test_map_thesis and test_map_all ?

tamtamz commented 3 years ago

test_map_all means that mAP is evaluated over all the verbs, while test_map_thesis means that mAP is evaluated over the target verbs, excluding stand, walk, etc. Note that the original evaluation code should be used for precise evaluation.