Is it a fair comparison with others if you use gt focal length in your evaluation?

JimmySuen / integral-human-pose

Integral Human Pose Regression

MIT License

471 stars 76 forks source link

Is it a fair comparison with others if you use gt focal length in your evaluation? #30

Open gdwei opened 5 years ago

gdwei commented 5 years ago

Hi, I wonder if it is a fair comparison with others if you use groundtruth focal length in your evaluation. This means you've known human size in test time and you do not consider the error introduced by the ambiguity of size ? Thanks.

John6333 commented 5 years ago

Agree with @gdwei. As far as I know, no other published 3D pose estimation methods before this work use the GT camera parameters or GT root depth, while this one used both.

victor-gz commented 5 years ago

Hi @gdwei, @John6333, where did you see they use the ground truth focal length? Could you point me to where it is in the code?

John6333 commented 5 years ago

Hi @vgonzalez88, please check this function, especially line623~637.

https://github.com/JimmySuen/integral-human-pose/blob/ad3f875bc05538da3471ef81484e23fad3e9c787/pytorch_projects/common_pytorch/dataset/hm36.py#L582