ziqima/Edit3D-Bench

This repository contains Edit3D-Bench, a 3D editing benchmark proposed in the paper Feedforward 3D Editing via Text-Steerable Image-to-3D.

Project page: https://glab-caltech.github.io/steer3d/ Code: https://github.com/ziqi-ma/Steer3D

The metadata/ directory stores metadata information (the source and target object guid, and editing text) for texture, removal, and addition - each in a separate .csv file. The data/ directory contains source images (of unedited object), glbs of both source and target objects, and TRELLIS latents of both source and target objects, each indexed with the object guid.

Sample Usage

This dataset is designed to benchmark 3D editing capabilities. To use it for evaluation with the associated Steer3D codebase, follow these steps:

1. Clone the dataset:

   git lfs install
   git clone https://huggingface.co/datasets/ziqima/Edit3D-Bench
   

2. Set up the Steer3D environment: The Steer3D model requires a specific environment. Refer to the Steer3D GitHub repository for the latest setup instructions. A typical setup involves:

   conda env create -f environment.yml
   conda activate steer3d
   

   Note that libraries kaolin, nvdiffrast, diffoctreerast, mip-splatting, and vox2seq might need manual installation. Please refer to this setup script from TRELLIS for installation of these dependencies.

3. Evaluate on the Benchmark (Texture Example): Once the dataset is cloned and the Steer3D environment is active, you can run evaluation scripts. First, ensure PYTHONPATH is set to the path of your Steer3D clone. Then, update the val_dataset path in configs/stage3_controlnet.json within the Steer3D repository to [path-to-Edit3D-Bench-clone]/metadata/texture.csv.

   python inference/inference_texture.py \
           --stage1_checkpoint [path-to-checkpoints]/stage1/base.pt \
           --stage1_config configs/stage1_controlnet.json \
           --stage2_controlnet_checkpoint [path-to-checkpoints]/stage2/controlnet.pt \
           --stage2_base_checkpoint [path-to-checkpoints]/stage2/base.pt \
           --stage2_config configs/stage2_controlnet.json \
           --output_dir visualizations/output \
           --num_examples 150 \
           --num_seeds 3 \
           --split val
   

Citation

If you find our work helpful, please cite using the following BibTeX entry:

@misc{ma2025feedforward3deditingtextsteerable,
      title={Feedforward 3D Editing via Text-Steerable Image-to-3D}, 
      author={Ziqi Ma and Hongqiao Chen and Yisong Yue and Georgia Gkioxari},
      year={2025},
      eprint={2512.13678},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2512.13678}, 
}