Learning Pixel-adaptive Multi-layer Perceptrons for Real-time Image Enhancement (ICCV 2025)

This repository is an official implementation of the paper "Learning Pixel-adaptive Multi-layer Perceptrons for Real-time Image Enhancement", ICCV, 2025.

Preparation

Dataset

Download FiveK (480p), FiveK(4K), PPR10K, LCDP dataset, and place them in your ./datasets.

Environment

• Python 3.10
• PyTorch 2.6.0

Installation

Download the code:

git clone https://github.com/LabShuHangGU/BPAM.git
cd bpam

Install the conda environment:

conda create -n bpam python=3.10
conda activate bpam

Install PyTorch:

# CUDA 12.4
pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu124

Install other requirements:

pip install -r requirements.txt

Training

• Refer to the training configuration files in ./options/train folder for detailed settings.

  CUDA_VISIBLE_DEVICES=0 python train.py -opt options/train/train_tone.yml
  CUDA_VISIBLE_DEVICES=0 python train.py -opt options/train/train_tone_4k.yml
  CUDA_VISIBLE_DEVICES=0 python train.py -opt options/train/train_fivek_pr.yml
  CUDA_VISIBLE_DEVICES=0 python train.py -opt options/train/train_ppr10k.yml
  CUDA_VISIBLE_DEVICES=0 python train.py -opt options/train/train_lcdp.yml
  

• ⚡️Though we have controlled the random seed, randomness still exists in training. We believe this may be caused by the function grid_sample, please refer to the note. To achieve the results presented in the paper, multiple training attempts may be required.

Testing

• Refer to the testing configuration files in ./options/test folder for detailed settings.

  # Evaluate the results
  CUDA_VISIBLE_DEVICES=0 python test.py -opt options/test/test_tone.yml
  CUDA_VISIBLE_DEVICES=0 python test.py -opt options/test/test_tone_4k.yml
  CUDA_VISIBLE_DEVICES=0 python test.py -opt options/test/test_fivek_pr.yml
  CUDA_VISIBLE_DEVICES=0 python test.py -opt options/test/test_ppr10k.yml
  CUDA_VISIBLE_DEVICES=0 python test.py -opt options/test/test_lcdp.yml
  
  # Evaluate the speed
  CUDA_VISIBLE_DEVICES=0 python test_speed.py -opt options/test/test_speed.yml
  

• (Optional) CUDA Implementation

  We also implement the CUDA extension for MLP to speed up.

  from models.archs.cpp_ext_interface import slice_function,apply_coefficent
  
  cd models/bilateral_slicing
  python setup.py install
  
  cd models/apply_coeff
  python setup.py install
  

​ 🪐 Replace the function MLP in the model with MLP_CUDA.

​ 🛸 The CUDA version can only be used for inference.