### Set up environment with setup_env.sh

Source: https://github.com/weijielyu/facelift/blob/main/README.md

Run this command to set up the environment for FaceLift. It is the only setup command provided in the README.

```bash
bash setup_env.sh
```

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

### Launch Gradio web interface

Source: https://github.com/weijielyu/facelift/blob/main/README.md

Launches the interactive Gradio web interface. After starting, open http://localhost:7860 in your browser, or use the provided public link if running on a server.

```bash
python gradio_app.py
```

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

### Run inference from command line

Source: https://github.com/weijielyu/facelift/blob/main/README.md

Runs inference on images in a directory. Use --input_dir to specify the input folder and --output_dir for results. Defaults are examples/ and outputs/ respectively.

```bash
python inference.py --input_dir examples/ --output_dir outputs/
```

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

### Train Gaussian reconstructor

Source: https://github.com/weijielyu/facelift/blob/main/README.md

Trains the Gaussian reconstructor (GS-LRM) using torchrun with 8 processes on 1 node. Uses the c10d rendezvous backend and requires the config configs/gslrm.yaml. Pre-trained weights from Objaverse data are used.

```bash
torchrun --nproc_per_node 8 --nnodes 1 \
    --rdzv_id ${JOB_UUID} --rdzv_backend c10d --rdzv_endpoint localhost:29500 \
    train_gslrm.py --config configs/gslrm.yaml
```

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

### Train multi-view diffusion model

Source: https://github.com/weijielyu/facelift/blob/main/README.md

Trains the multi-view diffusion model using Accelerate with an 8-GPU configuration. Requires the mvdiffusion/node_config/8gpu.yaml config file and the training config configs/mvdiffusion.yaml.

```bash
accelerate launch --config_file mvdiffusion/node_config/8gpu.yaml \
    train_diffusion.py --config configs/mvdiffusion.yaml
```

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

### GS-LRM data structure

Source: https://github.com/weijielyu/facelift/blob/main/README.md

Directory structure for GS-LRM training data. Place training and validation lists in data_gslrm_train.txt and data_gslrm_val.txt, and organize 32 multi-view images (RGBA, 512×512) under images/ along with opencv_cameras.json for camera parameters.

```text
data_sample/
├── gslrm/
│   ├── data_gslrm_train.txt           # Training data list
│   ├── data_gslrm_val.txt             # Validation data list
│   └── sample_000/
│       ├── images/
│       │   ├── cam_000.png            # Multi-view images (RGBA, 512×512)
│       │   ├── cam_001.png
│       │   ├── ...
│       │   └── cam_031.png            # 32 views total
│       └── opencv_cameras.json        # Camera parameters
```

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

### Multi-view diffusion data structure

Source: https://github.com/weijielyu/facelift/blob/main/README.md

Directory structure for multi-view diffusion training data. Place training and validation lists in data_mvdiff_train.txt and data_mvdiff_val.txt, and organize sample images as cam_000.png through cam_005.png (RGBA, 512×512).

```text
data_sample/
├── mvdiffusion/
│   ├── data_mvdiff_train.txt          # Training data list
│   ├── data_mvdiff_val.txt            # Validation data list
│   └── sample_000/
│       ├── cam_000.png                # Front view (RGBA, 512×512)
│       ├── cam_001.png                # Front-right view
│       ├── cam_002.png                # Right view
│       ├── cam_003.png                # Back view
│       ├── cam_004.png                # Left view
│       └── cam_005.png                # Front-left view
```