search

princeton-vl / pose-hg-train

Training and experimentation code used for "Stacked Hourglass Networks for Human Pose Estimation"
Other
575 stars 185 forks source link

Is the pre-trained model trained using both the FLIC AND MPII dataset? #44

Open neherh opened 6 years ago

neherh commented 6 years ago

Is the pre-trained model trained using both the FLIC AND MPII dataset? Or is it just pre-trained on one of the datasets? In your paper, you are unclear as to whether you trained on both or separately.

anewell commented 6 years ago

The pretrained model is trained only on MPII. Benchmarking for the paper is done by training and evaluating on a each dataset separately.

vicdu commented 6 years ago

The result in the paper is 90.09, but the recurring result is 88.78, where do you think the problem occurred?

anewell commented 6 years ago

If you're getting 88.78 on validation then that's probably fine. The performance on the official test set (what we report in the paper) is always different, and usually a little bit higher than validation performance in my experience.