Visual-AI / RegionDrag

The official repository for ECCV2024 paper "RegionDrag: Fast Region-Based Image Editing with Diffusion Models"
34 stars 3 forks source link

Evaluation #1

Open AsukaCamellia opened 2 months ago

AsukaCamellia commented 2 months ago

Thanks for such great work! Can I get the MD and LPIPS by just running the run_eval.py? Could you provide a more detailed description of training and evaluation in readme?

LuJingyi-John commented 1 month ago

Thank you so much for your interest in our work! You can use run_eval.py to evaluate our results. Our method is training-free and uses only a pretrained SD1-5 model. To evaluate:

  1. Install RegionDrag and download the DragBench datasets (SR and DR) by following the instructions in our README.
  2. Ensure your file structure matches:
RegionDrag/
├── assets/
├── utils/
├── drag_data/
│   ├── dragbench-dr/
│   └── dragbench-sr/
├── README.md
├── UI_GUIDE.md
├── requirements.txt
├── run_eval.py
└── ui.py
  1. Run the evaluation:
    • For DragBench-SR: python run_eval.py --data_dir drag_data/dragbench-sr/
    • For DragBench-DR: python run_eval.py --data_dir drag_data/dragbench-dr/
AsukaCamellia commented 1 month ago

Thank you so much for your interest in our work! You can use run_eval.py to evaluate our results. Our method is training-free and uses only a pretrained SD1-5 model. To evaluate:

  1. Install RegionDrag and download the DragBench datasets (SR and DR) by following the instructions in our README.
  2. Ensure your file structure matches:
RegionDrag/
├── assets/
├── utils/
├── drag_data/
│   ├── dragbench-dr/
│   └── dragbench-sr/
├── README.md
├── UI_GUIDE.md
├── requirements.txt
├── run_eval.py
└── ui.py
  1. Run the evaluation:

    • For DragBench-SR: python run_eval.py --data_dir drag_data/dragbench-sr/
    • For DragBench-DR: python run_eval.py --data_dir drag_data/dragbench-dr/

Thanks!

AsukaCamellia commented 1 month ago

Thank you so much for your interest in our work! You can use run_eval.py to evaluate our results. Our method is training-free and uses only a pretrained SD1-5 model. To evaluate:

  1. Install RegionDrag and download the DragBench datasets (SR and DR) by following the instructions in our README.
  2. Ensure your file structure matches:
RegionDrag/
├── assets/
├── utils/
├── drag_data/
│   ├── dragbench-dr/
│   └── dragbench-sr/
├── README.md
├── UI_GUIDE.md
├── requirements.txt
├── run_eval.py
└── ui.py
  1. Run the evaluation:

    • For DragBench-SR: python run_eval.py --data_dir drag_data/dragbench-sr/
    • For DragBench-DR: python run_eval.py --data_dir drag_data/dragbench-dr/

Hi, thank you for your assistance. I've got the results, but there are several points that need further attention.

  1. According to lines 26-32 in utils/cycle_sde.py, we need the IP-Adapter.
  2. If we chose SD 1.5, we should change the path (from 2.1 to 1.5) in lines 196-200 in utils/evaluator.py.
  3. It is imperative to install diffusers==0.27.2 specifically, as installing the latest version will result in peculiar errors.
LuJingyi-John commented 1 month ago

Thank you for your valuable feedback. Here are some explanations that may help: