README.md

Anything related to preprocessing for PALM-Net format

Preprocess in-the-wile images for PALM-Net personalization

Install hamer as per their instructions (or see installation guide here), then put images under hamer/images.

conda activate hamer
python demo.py --seq_name blah --batch_size=2  --full_frame --body_detector regnety # this will dump itw_hamer.npy

Unpack Hamer results:

python unpack_hamer.py --hpe_path ./itw_hamer.npy

MANO registration:

python register_mano.py --save_mesh --use_beta_loss

Install lang-segment-anything following their instructions. Extract hand masks:

conda activate lang-sam
cd lang-segment-anything
python demo.py
cd ..

Build datasets:

python build_dataset.py --rebuild --build_all

Convert InterHand2.6M data to PALM-Net format

Download InterHand2.6M annotations and put them in this structure:

./interhand/annotations/ # `annotations` is the downloaded folder from IH

Create InterHand2.6M folders:

python interhand2PALM-Net.py

Extract hand masks:

conda activate lang-sam
cd lang-segment-anything
python demo.py ## Change the glob path inside and use the keyword "foreground."
cd ..