Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Code for data preprocessing

See original GitHub issue

Thanks for the code again.

Would you also provide the code for data preprocessing? For example, how did you reformat the original H3.6m dataset and generate the .h5 files? How did you finetune the stack-hourglass model on H3.6m data?

That would be a great reference for manipulating other datasets. Thank you very much!

Issue Analytics

State:
Created 6 years ago
Comments:6 (3 by maintainers)

Top GitHub Comments

7reactions

una-dinosauriacommented, Sep 5, 2017

Here you can find the fine-tuned SH detections we used. You can replace them by the ones we provide by default and get slightly better results: https://drive.google.com/open?id=0BxWzojlLp259S2FuUXJ6aUNxZkE

2reactions

una-dinosauriacommented, Sep 5, 2017

Here is the link for the h5 files that we used to fine-tune Stacked Hourglass: https://drive.google.com/file/d/0BxWzojlLp259MS1obTA2QkNheDg/view?usp=sharing. I think this should work almost out of the box for fine-tuning, as it contains all the 2d data-to-image mappings in SH format.

(If you get the code from https://github.com/anewell/pose-hg-train, this goes under data/).

You’ll obviously also need the images too, which you should generate yourself from the videos using ffmpeg.

Here’s a gist used to generate these h5 files: like I said, not close to being releasable; only hoping it will help understand the process: https://gist.github.com/una-dinosauria/382e488d557af2b07e89bdeddeab1379