All commands start inside MonoFusion/preproc/ExoRecon. Raw takes live in ../../raw_data, processed assets in ../../data.

Register for EgoExo access at https://docs.ego-exo4d-data.org/, accept the license, then use the portal at https://visualize.ego4d-data.org/login to find the <take_uuid> you want.

egoexo -o ../../ \
  --uids <take_uuid> \
  --parts take_trajectory take_vrs takes -y
mv ../../raw_data/takes/* ../../raw_data/
rm -rf ../../raw_data/takes

Add the first valid RGB frame index to each take folder:

echo <first_frame_idx> > ../../raw_data/<take_id>/timestep.txt

conda env create -f egorecon.yml
conda activate egorecon
python -m pip install --upgrade pip
python -m pip install -e projectaria_tools_pkg

Edit LOCAL_BASE_PATH inside push_all_data.sh if paths differ, then:

./push_all_data.sh

This runs viewer_mps plus both preprocessing scripts for every take under ../../raw_data.

../../raw_data/<take_id>/
  aria01.vrs
  frame_aligned_videos/
    cam01.mp4 … cam04.mp4
  processed_frames/
    cam01/...
  undist_processed_frames/
    undist_cam01/...
  train.matrices.txt
  cam0X_train.matrices.txt
  trajectory/
    closed_loop_trajectory.csv
    gopro_calibs.csv
    Dy_train_meta.json
  timestep.txt
../../data/images/<take_id>/
  cam01/...
  cam02/...
  cam03/...
  cam04/...

If any folder is missing, re-run ./push_all_data.sh or the numbered scripts inside preprocess_scripts/ for that take.