DiffBIR

https://github.com/xpixelgroup/diffbir
Admin
DiffBIR is a blind image restoration pipeline that uses a generative diffusion prior to handle...
Tokens:8,921
Snippets:67
Trust Score:8
License:Apache-2.0
Update:8 months ago
Show doc for...
Tokens:
Raw
### Install DiffBIR and Setup Environment

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

This snippet shows how to clone the DiffBIR repository, create a conda environment, activate it, and install the necessary Python dependencies from the requirements file.

```bash
# Clone the repository
git clone https://github.com/XPixelGroup/DiffBIR.git
cd DiffBIR

# Create and activate conda environment
conda create -n diffbir python=3.10
conda activate diffbir
pip install -r requirements.txt
```

--------------------------------

### Start Stage 2 Training

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command starts the training process for Stage 2 of DiffBIR, utilizing the pretrained Stable Diffusion v2.1 model. It requires a configuration file and uses `accelerate launch` for distributed training.

```shell
accelerate launch train_stage2.py --config configs/train/train_stage2.yaml
```

--------------------------------

### Start Stage 1 Training

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command initiates the training process for Stage 1 of DiffBIR, which involves training a SwinIR model. It requires a configuration file specifying training parameters and uses `accelerate launch` for distributed training.

```shell
accelerate launch train_stage1.py --config configs/train/train_stage1.yaml
```

--------------------------------

### Install DiffBIR Project Dependencies

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This snippet outlines the steps to clone the DiffBIR repository, create a Conda environment, and install the necessary Python packages. It specifies Python 3.10 and provides an alternative for older PyTorch versions with xformers.

```shell
# clone this repo
git clone https://github.com/XPixelGroup/DiffBIR.git
cd DiffBIR

# create environment
conda create -n diffbir python=3.10
conda activate diffbir
pip install -r requirements.txt
```

--------------------------------

### Run Inference with Traditional DDIM Sampler (Bash)

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Example command to run DiffBIR inference using the traditional DDIM sampler. This demonstrates specifying a different sampler, step count, and start point type for comparison.

```bash
python -u inference.py \
    --task sr --version v2 \
    --sampler ddim \
    --steps 50 \
    --start_point_type cond \
    --input inputs/demo/bsr --output results/ddim
```

--------------------------------

### Run Inference with Fast Sampler (Bash)

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Example command to run the DiffBIR inference script for super-resolution using a fast sampler. It specifies the task, model version, sampler type, number of steps, and input/output directories.

```bash
python -u inference.py \
    --task sr --version v2.1 \
    --sampler edm_dpm++_3m_sde \
    --steps 10 \
    --input inputs/demo/bsr --output results/fast
```

--------------------------------

### Fetch and Process JSON Data (JavaScript)

Source: https://github.com/xpixelgroup/diffbir/blob/main/llava/eval/webpage/index.html

Fetches data from a 'data.json' file, parses it as JSON, and then populates models and questions. It also displays the current question. Errors during the fetch or processing are logged to the console.

```javascript
// Fetch the JSON file fetch('data.json') .then(response => response.json()) .then(json_data => { // Populate the models and questions. populateModels(json_data.models); populateQuestions(json_data.questions); displayQuestion(currentQuestionIndex); }).catch(error => console.error(error));
```

--------------------------------

### Blind Image Denoising (BID) Inference

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Remove noise from images while preserving and enhancing details using the denoising inference loop. Examples are provided for DiffBIR v2.1 and v2.

```bash
# DiffBIR v2.1 for image denoising
python -u inference.py \
    --task denoise \
    --upscale 1 \
    --version v2.1 \
    --captioner llava \
    --cfg_scale 8 \
    --noise_aug 0 \
    --input inputs/demo/bid \
    --output results/v2.1_demo_bid
```

```bash
# DiffBIR v2 for image denoising
python -u inference.py \
    --task denoise \
    --upscale 1 \
    --version v2 \
    --sampler spaced \
    --steps 50 \
    --captioner none \
    --pos_prompt '' \
    --neg_prompt 'low quality, blurry, low-resolution, noisy, unsharp, weird textures' \
    --cfg_scale 4.0 \
    --input inputs/demo/bid \
    --output results/v2_demo_bid \
    --device cuda --precision fp32
```

--------------------------------

### Blind Face Restoration (BFR) - Aligned Faces Inference

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Restore degraded face images that are pre-aligned using the specialized face restoration pipeline. Examples are provided for DiffBIR v2.1 and v2.

```bash
# DiffBIR v2.1 for aligned face restoration
python -u inference.py \
    --task face \
    --upscale 1 \
    --version v2.1 \
    --captioner llava \
    --cfg_scale 8 \
    --noise_aug 0 \
    --input inputs/demo/bfr/aligned \
    --output results/v2.1_demo_bfr_aligned
```

```bash
# DiffBIR v2 for aligned face restoration
python -u inference.py \
    --task face \
    --upscale 1 \
    --version v2 \
    --sampler spaced \
    --steps 50 \
    --captioner none \
    --pos_prompt '' \
    --neg_prompt 'low quality, blurry, low-resolution, noisy, unsharp, weird textures' \
    --cfg_scale 4.0 \
    --input inputs/demo/bfr/aligned \
    --output results/v2_demo_bfr_aligned \
    --device cuda --precision fp32
```

--------------------------------

### Train Stage 1 (SwinIR Cleaner)

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Train the stage-1 degradation removal model (SwinIR) on your own dataset. This involves generating file lists for training and validation, editing configuration files with your paths, and starting the training process using accelerate.

```bash
# 1. Generate file lists for training and validation sets
find /path/to/images -type f > files.list
shuf files.list > files_shuf.list
head -n 10000 files_shuf.list > train.list
tail -n +10001 files_shuf.list > val.list

# 2. Edit configs/train/train_stage1.yaml with your paths:
# - dataset.train.file_list: path/to/train.list
# - dataset.val.file_list: path/to/val.list
# - train.exp_dir: path/to/experiment/output

# 3. Start training with accelerate
accelerate launch train_stage1.py --config configs/train/train_stage1.yaml
```

--------------------------------

### Train Stage 2 (IRControlNet)

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Train the stage-2 ControlNet model for generating realistic details using diffusion. This requires downloading a pretrained Stable Diffusion v2.1 model, generating a training file list, editing configuration files, and starting training with accelerate.

```bash
# 1. Download pretrained Stable Diffusion v2.1
wget https://huggingface.co/stabilityai/stable-diffusion-2-1-base/resolve/main/v2-1_512-ema-pruned.ckpt --no-check-certificate

# 2. Generate training file list (same as stage 1)
find /path/to/images -type f > train.list

# 3. Edit configs/train/train_stage2.yaml with your paths:
# - train.sd_path: path/to/v2-1_512-ema-pruned.ckpt
# - train.swinir_path: path/to/stage1/checkpoint.pt
# - train.exp_dir: path/to/experiment/output
# - dataset.train.file_list: path/to/train.list

# 4. Start training with accelerate
accelerate launch train_stage2.py --config configs/train/train_stage2.yaml
```

--------------------------------

### Blind Face Restoration (BFR) - Unaligned/Whole Image Inference

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Restore faces within full scene images, automatically detecting and enhancing both faces and background. Examples are provided for DiffBIR v2.1 and v2.

```bash
# DiffBIR v2.1 for unaligned face + background restoration
python -u inference.py \
    --task face_background \
    --upscale 2 \
    --version v2.1 \
    --captioner llava \
    --cfg_scale 8 \
    --noise_aug 0 \
    --input inputs/demo/bfr/whole_img \
    --output results/v2.1_demo_bfr_unaligned
```

```bash
# DiffBIR v2 for unaligned face + background restoration
python -u inference.py \
    --task face_background \
    --upscale 2 \
    --version v2 \
    --sampler spaced \
    --steps 50 \
    --captioner none \
    --pos_prompt '' \
    --neg_prompt 'low quality, blurry, low-resolution, noisy, unsharp, weird textures' \
    --cfg_scale 4.0 \
    --input inputs/demo/bfr/whole_img \
    --output results/v2_demo_bfr_unaligned \
    --device cuda --precision fp32
```

--------------------------------

### Launch Gradio Web Interface

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Launch an interactive Gradio web UI for real-time image restoration. Options include specifying the captioner (LLaVA, RAM, or none) and controlling LLaVA quantization.

```bash
# Launch Gradio interface with LLaVA captioner (recommended)
python run_gradio.py --captioner llava

# For low-VRAM systems, use RAM captioner or disable captioning
python run_gradio.py --captioner ram
python run_gradio.py --captioner none

# Control LLaVA quantization (4-bit default, 8-bit, or 16-bit)
python run_gradio.py --captioner llava --llava_bit 8
```

--------------------------------

### Run DiffBIR Gradio Interface

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command initiates the Gradio web interface for DiffBIR. It includes an option to set the captioner to 'ram' or 'none' for users with low VRAM.

```shell
python run_gradio.py --captioner llava
```

--------------------------------

### Run Blind Image Super-Resolution Inference with DiffBIR v2

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command demonstrates how to perform Blind Image Super-Resolution using DiffBIR version 2. It specifies task, upscale factor, version, sampler, steps, captioner settings, prompts, CFG scale, input/output paths, device, and precision. The pretrained weights are automatically downloaded.

```shell
python -u inference.py \
--task sr \
--upscale 4 \
--version v2 \
--sampler spaced \
--steps 50 \
--captioner none \
--pos_prompt '' \
--neg_prompt 'low quality, blurry, low-resolution, noisy, unsharp, weird textures' \
--cfg_scale 4 \
--input inputs/demo/bsr \
--output results/v2_demo_bsr \
--device cuda --precision fp32
```

--------------------------------

### Run Blind Image Super-Resolution Inference with DiffBIR v2.1

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command shows how to perform Blind Image Super-Resolution using DiffBIR version 2.1. It configures the task, upscale factor, version, captioner, CFG scale, noise augmentation, and input/output paths. Pretrained weights are automatically downloaded.

```shell
python -u inference.py \
--task sr \
--upscale 4 \
--version v2.1 \
--captioner llava \
--cfg_scale 8 \
--noise_aug 0 \
--input inputs/demo/bsr \
--output results/v2.1_demo_bsr
```

--------------------------------

### Download Pretrained Models (Bash)

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Commands to download the DiffBIR pretrained models using wget. It provides URLs for different versions of the models, including the recommended v2.1.

```bash
# v2.1 (recommended)
wget https://huggingface.co/lxq007/DiffBIR-v2/resolve/main/DiffBIR_v2.1.pt

# v2
wget https://huggingface.co/lxq007/DiffBIR-v2/resolve/main/v2.pth

# v1 models
wget https://huggingface.co/lxq007/DiffBIR-v2/resolve/main/v1_general.pth
wget https://huggingface.co/lxq007/DiffBIR-v2/resolve/main/v1_face.pth
```

--------------------------------

### Run Blind Aligned-Face Restoration Inference with DiffBIR v2

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command illustrates Blind Aligned-Face Restoration using DiffBIR version 2. It sets the task, upscale factor, version, sampler, steps, captioner, prompts, CFG scale, input/output directories, device, and precision. Pretrained weights are automatically downloaded.

```shell
python -u inference.py \
--task face \
--upscale 1 \
--version v2 \
--sampler spaced \
--steps 50 \
--captioner none \
--pos_prompt '' \
--neg_prompt 'low quality, blurry, low-resolution, noisy, unsharp, weird textures' \
--cfg_scale 4.0 \
--input inputs/demo/bfr/aligned \
--output results/v2_demo_bfr_aligned \
--device cuda --precision fp32
```

--------------------------------

### Generate File Lists for Training Data

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This shell script demonstrates how to generate file lists for training and validation datasets. It uses `find` to collect image files, `shuf` to shuffle them, and `head`/`tail` to split them into training and validation sets.

```shell
# collect all iamge files in img_dir
find [img_dir] -type f > files.list
# shuffle collected files
shuf files.list > files_shuf.list
# pick train_size files in the front as training set
head -n [train_size] files_shuf.list > files_shuf_train.list
# pick remaining files as validation set
tail -n +[train_size + 1] files_shuf.list > files_shuf_val.list
```

--------------------------------

### Download Stable Diffusion v2.1 Checkpoint

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command downloads the pretrained Stable Diffusion v2.1 checkpoint, which is required for Stage 2 training. The `--no-check-certificate` flag is used to bypass SSL certificate verification.

```shell
wget https://huggingface.co/stabilityai/stable-diffusion-2-1-base/resolve/main/v2-1_512-ema-pruned.ckpt --no-check-certificate
```

--------------------------------

### Run Blind Aligned-Face Restoration Inference with DiffBIR v2.1

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command details how to perform Blind Aligned-Face Restoration with DiffBIR version 2.1. It configures the task, upscale factor, version, captioner, CFG scale, noise augmentation, and input/output paths. Pretrained weights are automatically downloaded.

```shell
python -u inference.py \
--task face \
--upscale 1 \
--version v2.1 \
--captioner llava \
--cfg_scale 8 \
--noise_aug 0 \
--input inputs/demo/bfr/aligned \
--output results/v2.1_demo_bfr_aligned
```

--------------------------------

### Run Custom Model Inference

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command demonstrates how to run inference with a custom-trained DiffBIR model. It requires specifying the path to the training configuration and the saved checkpoint. Key parameters include upscale factor, captioner, and noise augmentation.

```shell
python -u inference.py \
--upscale 4 \
--version custom \
--train_cfg [path/to/training/config] \
--ckpt [path/to/saved/checkpoint] \
--captioner llava \
--cfg_scale 8 \
--noise_aug 0 \
--input inputs/demo/bsr \
--output results/custom_demo_bsr
```

--------------------------------

### Run Blind Unaligned-Face Restoration Inference with DiffBIR v2

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command shows how to perform Blind Unaligned-Face Restoration using DiffBIR version 2. It specifies the task, upscale factor, version, sampler, steps, captioner, prompts, CFG scale, input/output directories, device, and precision. Pretrained weights are automatically downloaded.

```shell
python -u inference.py \
--task face_background \
--upscale 2 \
--version v2 \
--sampler spaced \
--steps 50 \
--captioner none \
--pos_prompt '' \
--neg_prompt 'low quality, blurry, low-resolution, noisy, unsharp, weird textures' \
--cfg_scale 4.0 \
--input inputs/demo/bfr/whole_img \
--output results/v2_demo_bfr_unaligned \
--device cuda --precision fp32
```

--------------------------------

### Run Blind Unaligned-Face Restoration Inference with DiffBIR v2.1

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command demonstrates Blind Unaligned-Face Restoration using DiffBIR version 2.1. It configures the task, upscale factor, version, captioner, CFG scale, noise augmentation, and input/output paths. Pretrained weights are automatically downloaded.

```shell
python -u inference.py \
--task face_background \
--upscale 2 \
--version v2.1 \
--captioner llava \
--cfg_scale 8 \
--noise_aug 0 \
--input inputs/demo/bfr/whole_img \
--output results/v2.1_demo_bfr_unaligned
```

--------------------------------

### Run Denoising Inference with DiffBIR v2.1

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command performs denoising using DiffBIR v2.1, featuring a different captioner and noise augmentation settings compared to v2. It specifies the input and output directories, along with other inference parameters.

```shell
python -u inference.py \
--task denoise \
--upscale 1 \
--version v2.1 \
--captioner llava \
--cfg_scale 8 \
--noise_aug 0 \
--input inputs/demo/bid \
--output results/v2.1_demo_bid
```

--------------------------------

### Enable Tiled Sampling for Inference

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

These arguments enable tiled sampling for various stages of the DiffBIR inference process, including the cleaner, VAE encoding/decoding, and the diffusion process itself. Tiled sampling is beneficial for super-resolution on low-VRAM GPUs.

```shell
[command...] \
# tiled inference for stage-1 model
--cleaner_tiled \
--cleaner_tile_size 256 \
--cleaner_tile_stride 128 \
# tiled inference for VAE encoding
--vae_encoder_tiled \
--vae_encoder_tile_size 256 \
# tiled inference for VAE decoding
--vae_decoder_tiled \
--vae_decoder_tile_size 256 \
# tiled inference for diffusion process
--cldm_tiled \
--cldm_tile_size 512 \
--cldm_tile_stride 256
```

--------------------------------

### Inference with Custom-Trained Model

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Run inference using your own trained IRControlNet checkpoint. This requires specifying the path to your training configuration file and the saved checkpoint.

```bash
# Inference with custom-trained model
python -u inference.py \
    --upscale 4 \
    --version custom \
    --train_cfg path/to/training/config.yaml \
    --ckpt path/to/saved/checkpoint.pt \
    --captioner llava \
    --cfg_scale 8 \
    --noise_aug 0 \
    --input inputs/demo/bsr \
    --output results/custom_demo_bsr
```

--------------------------------

### Run Denoising Inference with DiffBIR v2

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This command runs the denoising task using DiffBIR v2 with specified parameters. It includes settings for the model version, sampler, steps, prompts, and input/output paths. The device is set to CUDA for GPU acceleration and precision is set to fp32.

```shell
python -u inference.py \
--task denoise \
--upscale 1 \
--version v2 \
--sampler spaced \
--steps 50 \
--captioner none \
--pos_prompt '' \
--neg_prompt 'low quality, blurry, low-resolution, noisy, unsharp, weird textures' \
--cfg_scale 4.0 \
--input inputs/demo/bid \
--output results/v2_demo_bid \
--device cuda --precision fp32
```

--------------------------------

### Python API - DiffBIR Pipeline Usage

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Demonstrates how to use the DiffBIR project programmatically in Python. It covers loading the SwinIR cleaner and ControlLDM models, setting up the diffusion scheduler, creating the SwinIRPipeline, and processing an input image.

```python
import torch
import numpy as np
from PIL import Image
from omegaconf import OmegaConf
from diffbir.model import ControlLDM, SwinIR, Diffusion
from diffbir.pipeline import SwinIRPipeline
from diffbir.utils.common import instantiate_from_config, load_model_from_url
from diffbir.inference.pretrained_models import MODELS

device = "cuda"

# Load stage-1 cleaner model (SwinIR)
swinir = instantiate_from_config(OmegaConf.load("configs/inference/swinir.yaml"))
swinir.load_state_dict(load_model_from_url(MODELS["swinir_realesrgan"]))
swinir.eval().to(device)

# Load stage-2 model (ControlLDM)
cldm = instantiate_from_config(OmegaConf.load("configs/inference/cldm.yaml"))
sd_weight = load_model_from_url(MODELS["sd_v2.1_zsnr"])
cldm.load_pretrained_sd(sd_weight)
control_weight = load_model_from_url(MODELS["v2.1"])
cldm.load_controlnet_from_ckpt(control_weight)
cldm.eval().to(device)
cldm.cast_dtype(torch.float16)

# Load diffusion scheduler
diffusion = instantiate_from_config(OmegaConf.load("configs/inference/diffusion_v2.1.yaml"))
diffusion.to(device)

# Create pipeline
pipeline = SwinIRPipeline(swinir, cldm, diffusion, None, device)

# Load and process image
lq = Image.open("input.jpg").convert("RGB")
lq = lq.resize((lq.width * 4, lq.height * 4), Image.BICUBIC)  # Upscale for SR
lq_array = np.array(lq)[None]  # Add batch dimension

```

--------------------------------

### Run Restoration Pipeline (Python)

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Executes the DiffBIR restoration pipeline with specified parameters. It utilizes torch.no_grad() and torch.autocast for efficient computation. The pipeline takes low-quality image arrays as input and outputs restored images.

```python
with torch.no_grad(), torch.autocast(device, torch.float16):
    result = pipeline.run(
        lq_array,
        steps=10,
        strength=1.0,
        cleaner_tiled=False, cleaner_tile_size=256, cleaner_tile_stride=128,
        vae_encoder_tiled=False, vae_encoder_tile_size=256,
        vae_decoder_tiled=False, vae_decoder_tile_size=256,
        cldm_tiled=True, cldm_tile_size=512, cldm_tile_stride=256,
        pos_prompt="high quality, detailed",
        neg_prompt="low quality, blurry",
        cfg_scale=8.0,
        start_point_type="noise",
        sampler_type="edm_dpm++_3m_sde",
        noise_aug=0, rescale_cfg=False,
        s_churn=0, s_tmin=0, s_tmax=300, s_noise=1, eta=1, order=1
    )

# Save result
Image.fromarray(result[0]).save("output.png")
```

--------------------------------

### Enable Tiled Sampling for Inference

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

This command enables tiled sampling for all stages of the inference process to reduce VRAM usage. It specifies various tile sizes and strides for different components like the cleaner, VAE encoder/decoder, and ControlLDM.

```bash
python -u inference.py \
    --task sr \
    --upscale 4 \
    --version v2.1 \
    --captioner llava \
    --cfg_scale 8 \
    --input inputs/demo/bsr \
    --output results/tiled_output \
    --cleaner_tiled \
    --cleaner_tile_size 256 \
    --cleaner_tile_stride 128 \
    --vae_encoder_tiled \
    --vae_encoder_tile_size 256 \
    --vae_decoder_tiled \
    --vae_decoder_tile_size 256 \
    --cldm_tiled \
    --cldm_tile_size 512 \
    --cldm_tile_stride 256
```

--------------------------------

### Blind Image Super-Resolution (BSR) Inference

Source: https://context7.com/xpixelgroup/diffbir/llms.txt

Perform Blind Super-Resolution to upscale low-resolution images, restore details, and remove degradation artifacts. Two versions of DiffBIR are demonstrated: v2.1 (recommended, uses LLaVA for auto-captioning) and v2 (ECCV paper version, manual captioning).

```bash
# DiffBIR v2.1 (recommended) - uses LLaVA for auto-captioning
python -u inference.py \
    --task sr \
    --upscale 4 \
    --version v2.1 \
    --captioner llava \
    --cfg_scale 8 \
    --noise_aug 0 \
    --input inputs/demo/bsr \
    --output results/v2.1_demo_bsr
```

```bash
# DiffBIR v2 (ECCV paper version) - no auto-captioning
python -u inference.py \
    --task sr \
    --upscale 4 \
    --version v2 \
    --sampler spaced \
    --steps 50 \
    --captioner none \
    --pos_prompt '' \
    --neg_prompt 'low quality, blurry, low-resolution, noisy, unsharp, weird textures' \
    --cfg_scale 4 \
    --input inputs/demo/bsr \
    --output results/v2_demo_bsr \
    --device cuda --precision fp32
```

--------------------------------

### Citation for DiffBIR Research Paper

Source: https://github.com/xpixelgroup/diffbir/blob/main/README.md

This BibTeX entry provides the necessary details for citing the DiffBIR research paper in academic contexts. It includes title, authors, year, arXiv ID, and primary classification.

```bibtex
@misc{lin2024diffbir,
      title={DiffBIR: Towards Blind Image Restoration with Generative Diffusion Prior},
      author={Xinqi Lin and Jingwen He and Ziyan Chen and Zhaoyang Lyu and Bo Dai and Fanghua Yu and Wanli Ouyang and Yu Qiao and Chao Dong},
      year={2024},
      eprint={2308.15070},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}
```
Tokens:
Raw