https://github.com/JudgementH/RefAny3D
https://github.com/JudgementH/RefAny3D
Last synced: 4 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/JudgementH/RefAny3D
- Owner: JudgementH
- License: apache-2.0
- Created: 2026-01-27T14:19:35.000Z (4 months ago)
- Default Branch: master
- Last Pushed: 2026-01-29T15:30:54.000Z (4 months ago)
- Last Synced: 2026-01-30T04:32:22.557Z (4 months ago)
- Language: Python
- Size: 21.6 MB
- Stars: 1
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- awesome-diffusion-categorized - [Code
README
# RefAny3D: 3D Asset-Referenced Diffusion Models for Image Generation
## TODO List
- [x] Inference code and pretrained models.
- [x] Training code.
- [x] Training dataset.
## Quickstart
### Create environment
1. Clone the repository and create a conda environment:
```
git clone https://github.com/JudgementH/RefAny3D.git
conda create -n r3d python=3.10
conda activate r3d
```
2. Install requirements
```
pip install -r requirements.txt
```
### Inference
```
python demo.py \
--prompt "" \
--glb_path "" \
--output_dir ""
```
Alternatively, we provide some example scripts:
```
bash scripts/demo_chair.sh
bash scripts/demo_traffic_cone.sh
```
## Training
### Download Dataset
First, download the training [dataset](https://huggingface.co/datasets/JudgementH/RefAny3D-Dataset) and organize it in the appropriate directory structure.
### Configure Training
1. Edit the training configuration file `train/config/train_refany3d.yaml`:
- Update `json_file` path under `train.dataset` section (default: `"./dataset/meta_data.json"`)
- Update `data_root_path` path under `train.dataset` section (default: `"./dataset"`)
Example configuration:
```yaml
train:
dataset:
json_file: ""
data_root_path: ""
```
### Run Training
Execute the training script:
```
bash train/scripts/train_refany3d.sh
```
**Note:** The training script uses `accelerate` with 8 GPU processes by default. You can modify the number of processes in `train/scripts/train_refany3d.sh` by adjusting the `--num_processes` parameter.