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

How to deal with the keypoints which out of the image size #43

Open HansonXia opened 6 years ago

HansonXia commented 6 years ago

Hi, in MPII dataset, some keypoints are out of the image size, or the coordinates of some points are (0, 0), how to deal with these keypoints when generate heatmaps and calculate loss?

anewell commented 6 years ago

If coordinates are outside of the image frame, the ground truth heatmap consists of all zeros, the same loss (mean squared error) is applied.

ethanyanjiali commented 4 years ago

but if it's just one or two pixels off the border, we will still have a little bit gaussian remaining for that joint right? because the gaussian patch itself is at least 7x7.

HansonXia commented 4 years ago

but if it's just one or two pixels off the border, we will still have a little bit gaussian remaining for that joint right? because the gaussian patch itself is at least 7x7.

My approach is to set the heatmap to 0, and do not calculate the loss