I want to estimate the pose of people in a crowd situation using crowdpose.
So, I wanted to run the crowdpose model, but crowdpose's github only had evaluation code, not a pose estimation model.
After hearing that the crowdpose model was available in the "pytorch" branch of alphapose, I configured the environment and ran it.
"ubuntu 16.04/1080/cuda 9.0"
I ran the demo using the command below in the above environment and got the output json file.
I know that crowdpose has 14 key points, but the output json has 17 key points. Does the demo app fit the crowdpose model?
If --matching is used, can pose estimation be done with the crowdpose model? Also, is it possible to obtain an output image of the pose estimated by the crowdpose model using --matching --save_img?
I want to estimate the pose of people in a crowd situation using crowdpose. So, I wanted to run the crowdpose model, but crowdpose's github only had evaluation code, not a pose estimation model. After hearing that the crowdpose model was available in the "pytorch" branch of alphapose, I configured the environment and ran it.
"ubuntu 16.04/1080/cuda 9.0"
I ran the demo using the command below in the above environment and got the output json file.
"python3 demo.py --indir ${img_directory} --outdir examples/res --matching"
What I'm curious about is
I know that crowdpose has 14 key points, but the output json has 17 key points. Does the demo app fit the crowdpose model?
If --matching is used, can pose estimation be done with the crowdpose model? Also, is it possible to obtain an output image of the pose estimated by the crowdpose model using --matching --save_img?
I look forward to a quick reply. Thank you always