Wrong output format when predicting keypoints in preprocess

[nothonfz]

I preprocess my own data, and when predicting keypoints, the output is .jpg images instead of a .npy files.

I cloned a specific repository using the link provided in the Dockerfile, using the command git clone --recurse-submodules https://github.com/jiangwei221/mmpose.git.

Then, I created the open-mmlab env with python 3.8 and the following commands.

1. pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu121
2. pip install mmcv-full==1.5.0
3. cd mmpose
4. pip install -r requirements.txt
5. pip install -v -e .

I also downloaded the pretrained model by myself and put it in the mmpose.

Using this env, this command { python demo/bottom_up_img_demo.py configs/body/2d_kpt_sview_rgb_img/associative_embedding/coco/higherhrnet_w48_coco_512x512_udp.py higher_hrnet48_coco_512x512_udp-7cad61ef_20210222.pth --img-path /root/ml-neuman-main/preprocess/data/my_video/output/images --out-img-root /root/ml-neuman-main/preprocess/data/my_video/output/keypoints --kpt-thr=0.3 --pose-nms-thr=0.9 } can be run successfully.

Howerve, when I ran the command bash -i run.sh, all the steps can run successfully but the tenth step. the output are like this.

And the error from the tenth step I looked up the output in keypoints

the output is .jpg images instead of a .npy files.

I looked up the code in /mmpose/demo/bottom_up_img_demo.py and found that save operation only exists in method vis_pose_result

I steped into the method vis_pose_result. And found that no code to save the result as .npy files.

Has anyone encountered the same issue? How can I save the result as .npy file with correct format?

[jiangwei221]

You need to checkout to the docker branch. https://github.com/jiangwei221/mmpose/tree/docker This branch has the dump code.

[jiangwei221]

https://github.com/jiangwei221/mmpose/blob/8b788f93200ce6485e885da0c736f114e4de8eaf/demo/bottom_up_img_demo.py#L127-L131

[nothonfz]

Thanks！