UniDG-Defect-Und-Reward

Defect-Und-Reward checkpoint for UniDG (UniDG: Unified Defect Generation via Defect-Context Editing, NeurIPS 2026), fine-tuned from Qwen/Qwen3-VL-4B-Instruct with LLaMA-Factory on 13,786 curated defect-editing evaluation conversations.

[Paper] [Code] [UniDG-SFT-LoRA] [UniDG-RFT-LoRA]

Overview

During Consistency-RFT (Flow-GRPO), Defect-Und-Reward evaluates each synthesized defect image alongside the reference image, reference mask, and target mask across four dimensions (0โ€“5 scale):

  • Reference Adherence
  • Consistency
  • Reasonability
  • Overall Quality

Usage in UniDG Consistency-RFT

Download this repository and set reward_model in train/rft/config.yaml:

hf download retofan23333/UniDG-Defect-Und-Reward-Release --local-dir checkpoints/UniDG-Defect-Und-Reward-Release

Citation

@inproceedings{fan2026unidg,
  title={UniDG: Unified Defect Generation via Defect-Context Editing},
  author={Fan, Yuanting and Liu, Jun and Gao, Bin-Bin and Chen, Xiaochen and Lin, Yuhuan and Dai, Zhewei and Zhan, Jiawei and Wang, Chengjie},
  booktitle={NeurIPS},
  volume={39},
  year={2026},
  publisher={Curran Associates, Inc.}
}
Downloads last month
32
Safetensors
Model size
4B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for retofan23333/UniDG-Defect-Und-Reward-Release

Finetuned
(574)
this model

Paper for retofan23333/UniDG-Defect-Und-Reward-Release