python -m venv env
source activate env/bin/activate
bash install.sh

# Step 1: Download data from: https://graphics.tu-bs.de/people-snapshot
# Step 2: Preprocess using our script
python scripts/peoplesnapshot/preprocess_PeopleSnapshot.py --root <PATH_TO_PEOPLESNAPSHOT> --subject male-3-casual
# Step 3: Download SMPL from: https://smpl.is.tue.mpg.de/ and place the model in ./data/SMPLX/smpl/
# └── SMPLX/smpl/
#         ├── SMPL_FEMALE.pkl
#         ├── SMPL_MALE.pkl
#         └── SMPL_NEUTRAL.pkl

Quickly learn and animate an avatar with bash ./bash/run-demo.sh

Here we use the in the wild video provided by Neuman as an example:

1. create a yaml file specifying the details about the sequence in ./confs/dataset/. In this example it's provided in ./confs/dataset/neuman/seattle.yaml.
2. download the data from Neuman's Repo, and run cp <path-to-neuman-dataset>/seattle/images ./data/custom/seattle/
3. run the bash script bash scripts/custom/process-sequence.sh ./data/custom/seattle neutral to preprocess the images, which
   • uses Openpose to estimate the 2D keypoints,
   • uses Segment-Anything to segment the scene
   • uses ROMP to estimate camera and smpl parameters
4. run the bash script bash ./bash/run-neuman-demo.sh to learn an avatar

And you can animate the avatar easily:

We would like to acknowledge the following third-party repositories we used in this project:

• [Tinycuda-nn]
• [Openpose]
• [ROMP]
• [Segment-anything]

Besides, we used code from:

• [Anim-NeRf]
• [SelfRecon]
• [lpips]
• [SMPLX]
• [pytorch3d]

We are grateful to the developers and contributors of these repositories for their hard work and dedication to the open-source community. Without their contributions, our project would not have been possible.

Please also check out our related projects!

• [Fast-SNARF]
• [Vid2Avatar]

@article{jiang2022instantavatar,
  author    = {Jiang, Tianjian and Chen, Xu and Song, Jie and Hilliges, Otmar},
  title     = {InstantAvatar: Learning Avatars from Monocular Video in 60 Seconds},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  month     = {June},
  year      = {2023},
}