human-pose-estimation.pytorch
human-pose-estimation.pytorch copied to clipboard
Question about flow-based propagation.
Hi, I recently try to re-implement the flow-based propagation section in your paper. I use the FlowNet2S you mentioned, and I get the flow information between two adjacent frames on posetrack2018 val dataset. The question is the inference size must be divisible by 64. Therefore, the shape of flow is not corresponding to the shape of images. In order to get the same results reported, I wonder how to solve this problem and the inference time for "With Joint Propagation" on posetrack2018 val dataset. Thanks a lot!