### Download assets via gagavatar-assets

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Downloads assets with explicit upstream sources and verifies known sizes and SHA-256 hashes. FLAME and tracker resources are not downloaded automatically due to separate setup or license requirements.

```bash
gagavatar-assets download --root assets
```

--------------------------------

### Install modified 3DGS renderer

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Clones, installs, and removes the modified 3DGS renderer. This version carries 32-dim features and uses a different package name to avoid conflicts.

```bash
git clone --recurse-submodules git@github.com:xg-chu/diff-gaussian-rasterization.git
pip install ./diff-gaussian-rasterization
rm -rf ./diff-gaussian-rasterization
```

--------------------------------

### Build resources for GAGAvatar_track

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Prepares resources for the GAGAvatar_track component.

```bash
cd core/libs/GAGAvatar_track
bash ./build_resources.sh
```

--------------------------------

### Build resources with build_resources.sh

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Builds required resources for GAGAvatar.

```bash
bash ./build_resources.sh
```

--------------------------------

### Create and activate Conda environment

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Creates a Conda environment from the provided environment.yml file and activates it.

```bash
conda env create -f environment.yml
conda activate GAGAvatar
```

--------------------------------

### Run inference with tracked video driver

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Runs inference driven by a tracked video sequence.

```bash
python inference.py -d ./demos/drivers/obama -i ./demos/examples/1.jpg
```

--------------------------------

### Run inference with image driver

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Runs inference driven by another image. This tracks the images online, which is slow.

```bash
python inference.py -d ./demos/examples/2.jpg -i ./demos/examples/1.jpg
```

--------------------------------

### Run inference with tracked image_lmdb driver

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Runs inference driven by a tracked image_lmdb dataset.

```bash
python inference.py -d ./demos/drivers/vfhq_demo -i ./demos/examples/1.jpg
```

--------------------------------

### optim.pkl structure

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Shows the structure of optim.pkl, containing keys like '000000_0' with sub-keys for bbox, shapecode, expcode, posecode, eyecode, and transform_matrix.

```text
optim.pkl:
- dict_keys(['000000_0', …])
    - "000000_0": dict_keys(['bbox', 'shapecode', 'expcode', 'posecode', 'eyecode', 'transform_matrix'])
```

--------------------------------

### Clone GAGAvatar repository

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Clones the GAGAvatar repository with all submodules and changes into the project directory.

```bash
git clone --recurse-submodules git@github.com:xg-chu/GAGAvatar.git
cd GAGAvatar
```

--------------------------------

### Train command

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Command to train the model with the specified config and dataset.

```bash
python train.py --config gaga --dataset vfhq
```

--------------------------------

### dataset.json structure

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Shows the structure of dataset.json, which should contain train, val, and test lists of keys from img_lmdb and optim.pkl.

```json
dataset.json: {
    "train": ["000000_0", "000000_5", ..., '001384_654'],
    "val":   ["015209_0", ..., "015218_7"], 
    "test":  ["015203_0", ..., "015252_139"]
}
```

--------------------------------

### img_lmdb structure

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

Shows the expected structure of the img_lmdb, where keys are video and frame IDs (e.g., '015252_99') mapping to image payloads.

```text
img_lmdb:
    '015252_99' : image payload
```

--------------------------------

### BibTeX citation

Source: https://github.com/xg-chu/gagavatar/blob/main/README.md

BibTeX entry for citing the GAGAvatar paper.

```bibtex
@inproceedings{
    chu2024gagavatar,
    title={Generalizable and Animatable Gaussian Head Avatar},
    author={Xuangeng Chu and Tatsuya Harada},
    booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems},
    year={2024},
    url={https://openreview.net/forum?id=gVM2AZ5xA6}
}
```