PanoVGGT

https://github.com/yijingguo-june/panovggt
Admin
PanoVGGT is a feed-forward, geometry-aware Transformer framework designed for globally consistent 3D...
Tokens:1,215
Snippets:18
Trust Score:5.9
License:MIT
Update:2 months ago
Show doc for...
Tokens:
Raw
### Install PanoVGGT Environment

Source: https://github.com/yijingguo-june/panovggt/blob/main/README.md

Commands to clone the repository, create a virtual environment, and install required PyTorch and project dependencies.

```bash
git clone https://github.com/YijingGuo-June/PanoVGGT.git
cd PanoVGGT

# 2. Create and activate a virtual environment (Recommended)
conda create -n panovggt python=3.11 -y
conda activate panovggt

# 3. Install PyTorch and related packages
# Note: The command below is for CUDA 12.4. You can adjust the versions 
# based on your local CUDA environment if needed.
pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 xformers==0.0.28.post3 --index-url https://download.pytorch.org/whl/cu124

# 4. Install the remaining dependencies
pip install -r requirements.txt
```

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

### Launch Gradio Web Interface

Source: https://github.com/yijingguo-june/panovggt/blob/main/README.md

Command to start the local Gradio application using pre-trained weights and a configuration file.

```bash
python app.py \
  --config training/config/default.yaml \
  --checkpoint checkpoints/model.pt \
  --device cuda \
  --port 7860
```

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

### Launch Distributed Training

Source: https://github.com/yijingguo-june/panovggt/blob/main/README.md

Command to initiate training or fine-tuning using the distributed launch script.

```bash
TORCH_NCCL_ASYNC_ERROR_HANDLING=1 torchrun --standalone --nproc_per_node=8 training/launch.py
```

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

### Download Model Weights

Source: https://github.com/yijingguo-june/panovggt/blob/main/README.md

Commands to download high-resolution and low-resolution pre-trained model checkpoints into the local checkpoints directory.

```bash
# Create a checkpoints directory
mkdir -p checkpoints

# Download the model weights
wget https://huggingface.co/YijingGuo/PanoVGGT/resolve/main/model.pt -O checkpoints/model.pt
```

```bash
wget https://huggingface.co/YijingGuo/PanoVGGT/resolve/main/model_lowres.pt -O checkpoints/model_lowres.pt
```

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

### Project Directory Structure

Source: https://github.com/yijingguo-june/panovggt/blob/main/README.md

Overview of the repository file layout.

```text
.
|-- app.py                 # Gradio demo entry
|-- panovggt/              # Core model, layers, geometry, projection
|-- training/              # Training configs, datasets, trainer, utils
|-- evaluation/            # Evaluation scripts and metrics
|-- examples/              # Example panoramic sequences
`-- requirements.txt       # Python dependencies
```

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

### Evaluate PanoVGGT model

Source: https://github.com/yijingguo-june/panovggt/blob/main/README.md

Run the evaluation script to reproduce quantitative results for geometry and pose estimation.

```bash
python evaluation/eval_allpano.py
```

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

### Cite PanoVGGT in BibTeX

Source: https://github.com/yijingguo-june/panovggt/blob/main/README.md

Use this BibTeX entry to cite the PanoVGGT research paper in academic publications.

```bibtex
@article{guo2026panovggt,
  title={PanoVGGT: Feed-Forward 3D Reconstruction from Panoramic Imagery},
  author={Guo, Yijing and Chao, Mengjun and Wang, Luo and Zhao, Tianyang and Dai, Haizhao and Zhang, Yingliang and Yu, Jingyi and Shi, Yujiao},
  journal={arXiv preprint arXiv:2603.17571},
  year={2026}
}
```
Tokens:
Raw