Repository files navigation

📃Paper • 🖼Dataset • 🌐中文博客 • 🤗HF Repo • 🐦Twitter

🔥🔥News!2024/12/31: We released thenext generation of model,VisionReward, which is a fine-grained and multi-dimensional reward model for stable RLHF for visual generation (text-to-image / text-to-video)!

🔥News!2023/9/22: The paper of ImageReward is accepted by NeurIPS 2023!

ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation

ImageReward is the first general-purpose text-to-image human preference RM, which is trained on in total137k pairs of expert comparisons, outperforming existing text-image scoring methods, such as CLIP (by 38.6%), Aesthetic (by 39.6%), and BLIP (by 31.6%), in terms of understanding human preference in text-to-image synthesis.

Additionally, we introduce Reward Feedback Learning (ReFL) for direct optimizing a text-to-image diffusion model using ImageReward. ReFL-tuned Stable Diffusion wins against untuned version by 58.4% in human evaluation.

Both ImageReward and ReFL are all packed up to Pythonimage-reward package now!

Tryimage-reward package in only 3 lines of code for ImageReward scoring!

# pip install image-rewardimportImageRewardasRMmodel=RM.load("ImageReward-v1.0")rewards=model.score("<prompt>", ["<img1_obj_or_path>","<img2_obj_or_path>", ...])

Tryimage-reward package in only 4 lines of code for ReFL fine-tuning!

# pip install image-reward# pip install diffusers==0.16.0 accelerate==0.16.0 datasets==2.11.0fromImageRewardimportReFLargs=ReFL.parse_args()trainer=ReFL.Trainer("CompVis/stable-diffusion-v1-4","data/refl_data.json",args=args)trainer.train(args=args)

If you findImageReward's open-source effort useful, please 🌟 us to encourage our following developement!

• ImageReward
  • Quick Start
    • Install Dependency
    • Example Use
  • ReFL
    • Install Dependency
    • Example Use
  • Demos of ImageReward and ReFL
  • Training code for ImageReward
  • Integration into Stable Diffusion Web UI
    • Features
      • Score generated images and append to image information
        • Usage
        • Demo video
      • Automatically filter out images with low scores
        • Usage
        • Demo video
      • View the scores of images that have been scored
        • Usage
        • Example
      • Other Features
        • Memory Management
    • FAQ
  • Reproduce Experiments in Table 1
  • Reproduce Experiments in Table 3
  • Citation

We have integrated the whole repository to a single python packageimage-reward. Following the commands below to prepare the environment:

# Clone the ImageReward repository (containing data for testing)git clone https://github.com/THUDM/ImageReward.gitcd ImageReward# Install the integrated package `image-reward`pip install image-reward

We provide example images in theassets/images directory of this repo. The example prompt is:

a painting of an ocean with clouds and birds, day time, low depth field effect

Use the following code to get the human preference scores from ImageReward:

importosimporttorchimportImageRewardasRMif__name__=="__main__":prompt="a painting of an ocean with clouds and birds, day time, low depth field effect"img_prefix="assets/images"generations= [f"{pic_id}.webp"forpic_idinrange(1,5)]img_list= [os.path.join(img_prefix,img)forimgingenerations]model=RM.load("ImageReward-v1.0")withtorch.no_grad():ranking,rewards=model.inference_rank(prompt,img_list)# Print the resultprint("\nPreference predictions:\n")print(f"ranking ={ranking}")print(f"rewards ={rewards}")forindexinrange(len(img_list)):score=model.score(prompt,img_list[index])print(f"{generations[index]:>16s}:{score:.2f}")

The output should be like as follow (the exact numbers may be slightly different depending on the compute device):

Preference predictions:ranking = [1, 2, 3, 4]rewards = [[0.5811622738838196], [0.2745276093482971], [-1.4131819009780884], [-2.029569625854492]]          1.webp: 0.58          2.webp: 0.27          3.webp: -1.41          4.webp: -2.03

pip install diffusers==0.16.0 accelerate==0.16.0 datasets==2.11.0

We provide example dataset for ReFL in thedata/refl_data.json of this repo. Run ReFL as following:

bash scripts/train_refl.sh

1. Download data: 🖼Dataset.

2. Make dataset.

cd trainpython src/make_dataset.py

3. Set training config:train/src/config/config.yaml

4. One command to train.

bash scripts/train_one_node.sh

We have developed acustom script to integrate ImageReward into SD Web UI for a convenient experience.

The script is located atsdwebui/image_reward.py in this repository.

Theusage of the script is described as follows:

1. Install: put the custom script into thestable-diffusion-webui/scripts/ directory
2. Reload: restart the service, or click the"Reload custom script" button at the bottom of the settings tab of SD Web UI. (If the button can't be found, try clicking the"Show all pages" button at the bottom of the left sidebar.)
3. Select: go back to the"txt2img"/"img2img" tab, and select"ImageReward - generate human preference scores" from the "Script" dropdown menu in the lower left corner.
4. Run: the specific usage varies depending on the functional requirements, as described in the"Features" section below.

1. Do not check the "Filter out images with low scores" checkbox.
2. Click the"Generate" button to generate images.
3. Check the ImageReward at thebottom of the image informationbelow the gallery.

1. Check the"Filter out images with low scores" checkbox.
2. Enter the score lower limit in"Lower score limit". (ImageReward roughly follows the standard normal distribution, with a mean of 0 and a variance of 1.)
3. Click the"Generate" button to generate images.
4. Images with scores below the lower limit will be automatically filtered out andwill not appear in the gallery.
5. Check the ImageReward at thebottom of the image informationbelow the gallery.

1. Upload the scored image file in the"PNG Info" tab
2. Check the image information on the right with the score of the image at thebottom.

• ImageReward model will not be loadeduntil first script run.
• "Reload UI" will not reload the model nor unload it, butreuses the currently loaded model (if it exists).
• A"Unload Model" button is provided to manually unload the currently loaded model.

Note thatSD Web UI has two ways to set up its Python environment:

• If youlaunch with python launch.py, Web UI will use the Python environment found in yourPATH (in Linux, you can check its exact path withwhich python).
• If youlaunch with a script like webui-user.bat, Web UI creates a new venv environment in the directorystable-diffusion-webui\venv.
  • Generally, you need some other operations to activate this environment. For example, in Windows, you need to enter the stable-diffusion-webui\venv\Scripts directory, run activate or activate.bat (if you are using cmd) or activate.ps1 (if you are using PowerShell) from .
  • If you see the prompt (venv) appear at the far left of the command line, you have successfully activated venv created by the SD Web UI.

After activating the right Python environment, just do what you want to do true to form.

Note: The experimental results are produced in an environment that satisfies:

• (NVIDIA) Driver Version: 515.86.01
• CUDA Version: 11.7
• torch Version: 1.12.1+cu113According to our own reproduction experience, reproducing this experiment in other environments may cause the last decimal place to fluctuate, typically within a range of ±0.1.

Run the following script to automatically download data, baseline models, and run experiments:

bash ./scripts/test-benchmark.sh

Then you can check the results inbenchmark/results/ or the terminal.

If you want to check the raw data files individually:

• Test prompts and corresponding human rankings for images are located inbenchmark/benchmark-prompts.json.
• Generated outputs for each prompt (originally fromDiffusionDB) can be downloaded fromHugging Face orTsinghua Cloud.
  • Each<model_name>.zip contains a directory of the same name, in which there are in total 1000 images generated from 100 prompts of 10 images each.
  • Every<model_name>.zip should be decompressed intobenchmark/generations/ as directory<model_name> that contains images.

Run the following script to automatically download data, baseline models, and run experiments:

bash ./scripts/test.sh

If you want to check the raw data files individually:

• Test prompts and corresponding human rankings for images are located indata/test.json.
• Generated outputs for each prompt (originally fromDiffusionDB) can be downloaded fromHugging Face orTsinghua Cloud. It should be decompressed todata/test_images.

@inproceedings{xu2023imagereward,  title={ImageReward: learning and evaluating human preferences for text-to-image generation},  author={Xu, Jiazheng and Liu, Xiao and Wu, Yuchen and Tong, Yuxuan and Li, Qinkai and Ding, Ming and Tang, Jie and Dong, Yuxiao},  booktitle={Proceedings of the 37th International Conference on Neural Information Processing Systems},  pages={15903--15935},  year={2023}}