search

amundra15 / livehand

Official PyTorch implementation of the ICCV '23 paper LiveHand
MIT License
59 stars 5 forks source link

How to choose annot_subset in data preprocessing #10

Closed Eydcao closed 7 months ago

Eydcao commented 10 months ago

Hi dear authors,

Fantastic paper! I am trying to follow up but encountering problems while preparing data from the 2.6M InterHand dataset.

In your livehand/preprocessing/InterHands2.6M/prepare_dataset.py, there is:

annot_subset = 'all'  # 'human'

This leads to InterHand2.6M_5fps_batch1/annotations/all/, i.e., a /all/ subfolder. However, this is inconsistent with the Google Drive you provided, where there are no annot_subset folders. Indeed, using your provided structure without a subfolder, the script reports a file-not-found error.

I tried creating this all folder myself and moving all test/train/valid folders inside, but then received warnings like below:

WARNING: You are using a MANO model with only 10 shape coefficients.
WARNING: You are using a MANO model with only 10 shape coefficients.
Fix shapedirs bug of MANO
WARNING: You are using a MANO model with only 10 shape coefficients.
WARNING: You are using a MANO model with only 10 shape coefficients.

I am unsure if this is expected and am wondering what annot_subset means and how I should set it correctly?

Best,

amundra15 commented 7 months ago

Hi, I think you have created the folders correctly. The warnings you see are from the MANO class and are expected.

In case there are still some issues with the pre-processing, you can try using the pre-processed data I have uploaded today.