When I was training the scoring net, model/pose_coco/pose_dekr_hrnetw32_coco.pth in the command could not be found. It was a file, so I changed it to ‘output/coco_kpt/hrnet_dekr/w32_4x_reg03_bs10_512_adam_lr1e-3_coco_x140/final_state0.pth.tar’. is this command correct, because he reported an error that the picture size is inconsistent? I can't find how to change the picture size. Please help me solve it~

HRNet / DEKR

This is an official implementation of our CVPR 2021 paper "Bottom-Up Human Pose Estimation Via Disentangled Keypoint Regression" (https://arxiv.org/abs/2104.02300)

MIT License

438 stars 76 forks source link

When I was training the scoring net, model/pose_coco/pose_dekr_hrnetw32_coco.pth in the command could not be found. It was a file, so I changed it to ‘output/coco_kpt/hrnet_dekr/w32_4x_reg03_bs10_512_adam_lr1e-3_coco_x140/final_state0.pth.tar’. is this command correct, because he reported an error that the picture size is inconsistent? I can't find how to change the picture size. Please help me solve it~ #29

Closed LiPan123456789 closed 2 years ago

LiPan123456789 commented 2 years ago

Gengzigang commented 2 years ago

You should use batchsize=1 in the inference process.

LiPan123456789 commented 2 years ago

The problem was solved successfully, but I want to know why.

Gengzigang commented 2 years ago

The input size of each image during inference is different.

LiPan123456789 commented 2 years ago

Thank you very much.