search

mlpc-ucsd / PRTR

(CVPR 2021) PRTR: Pose Recognition with Cascade Transformers
Apache License 2.0
141 stars 29 forks source link

Person-detection Transformer in two-stage variant #3

Closed Charleshhy closed 3 years ago

Charleshhy commented 3 years ago

Hi,

Thanks for the nice work, I am just get a little bit confused about the implementation. The proposed person-detection transformer is not used in the two-stage variant. Instead, person detection results from another detector and gt bounding boxes are used in COCO and MPII, respectively.

Am I understanding it right? If the proposed person-detection transformer is used, can you point it out where and how? Thanks a lot!

Regards, Charles

likenneth commented 3 years ago

Hi Charles, Thanks for your interest in PRTR. You are half-correct here.
MPII dataset contains only single-person images, which does not require bbox. For COCO, we used ground truth person bounding boxes for training but finetuned person detection Transformer outputs in the evaluation stage. Details of the finetuning process can be found at Person-detection Transformer finetuning of 4.1 Experiment Setup in the paper.

Charleshhy commented 3 years ago

Thanks a lot for the explanation! I will close the issue then.