A question about the data used in PoseFlow/deepmatching.py

[lc4321]

Hi, could you tell me the detailed name of the data used in PoseFlow/deepmatching.py, I donn't know which the data you used that satisfied the condition of "len(a.split("/")) == 9". When I debug the code, the value of vidnames is empty.

[lc4321]

I cann't find the document of alpha-pose-results-{}.json, is it just the alpha-pose-results-forvis.json? When I execute the run.sh, both alpha-pose-results.json and alpha-pose-results-forvis.json are generated in Alphapose/examples/results/Pose. 3Q!

[YuliangXiu]

You can not run this code directly without downloading PoseTrack challenge dataset, downloading introduction can be found in README. Besides, I have updated the new PoseFlow codes, adding one alpha-pose-results-sample.json for describing the structure of json file.

[lc4321]

Thank you for your reply.
The error that "cann't find json" is escaped by replacing "json" with "mat" in This line. But line
234 still needs the .json, The documents I loaded include only .mat instead of .json. Where can I load it? Or the .json needs to be generated from .mat by myself?

[YuliangXiu]

Please read README file carefully...

[lc4321]

Thank you! I have converted the annotation file from Mat to Json with your help. Finanlly, I got the xx_mpii_relpath_5sec_testsub.json file after executing the tracker.py file, How to show the tracking effect for a video by using the xx_mpii_relpath_5sec_testsub.json file?

[YuliangXiu]

tracking changes the person id in every frame.

[lc4321]

If I use the images converted from myself video, how to make the Mat file which will be converted to Json file?3Q!

[lc4321]

Sorry to bother you again. I find that content of point:[] in the xx_mpii_relpath_5sec_testsub.json file generated by tracker.py is empty,. and the reason is that all the pose_score<drop. Is that normal?

[YuliangXiu]

I did not test my code on the user-defined dataset, so I think it is better to write your own script to convert mat to json(refer to the data format of current generated json).

I do not think it is normal. Where do the scores come from, AlphaPose? Maybe you can write some visualization code to render skeleton results to make sure your results are reasonable.

[lc4321]

The "pose_scoreI" in my file is come from Alphapose. All the values of "pose_scoreI" are between 0~1, I still cann't find where the problem is. I have adjusted the value of drop to 0.2, then the content of point:[] in the final xx_mpii_relpath_5sec_testsub.json file is not empty. The tracking effect is better than ever.

[YuliangXiu]

Yes, I think this is my fault, I do some post-processing after pose estimation, so my pose score is not between 0 and 1, you can tune the drop parameter according to your own dataset.

[lc4321]

Can I find your post-processing job in the AlphaPose project cloned from this github? Or you have not upload this job?

[YuliangXiu]

I did not include this post-processing code into AlphaPose because this module has a minor impact on final results and the source code is in mess, so you can just tune parameters on your current results.

[YuliangXiu]

PoseFlow(General Version) has already been released, now you can do pose tracking on any private dataset, the new version also supports pose tracking results visualization.