Skip to content

INSTALL.md

Latest commit

 

History

History
67 lines (48 loc) · 1.56 KB

INSTALL.md

File metadata and controls

67 lines (48 loc) · 1.56 KB

Installation

The codebase consists of two independent components:

  • Dataset Generator (main/): Simulates event-based lens flare data
  • E-DeflareNet (Unet_main/): Neural network for flare removal

Prerequisites

  • Python 3.9-3.10
  • CUDA-capable GPU (recommended)
  • Git

Step 1: Dataset Generator Setup

cd main

# Install dependencies
pip install -r requirements.txt

# Verify installation
python -c "import torch, h5py, cv2; print('✅ Dataset generator ready')"

Step 2: E-DeflareNet Setup

cd ../Unet_main

# Create conda environment
conda create -n event_deflare python=3.9 \
  pytorch torchvision pytorch-cuda=12.1 \
  pytorch-3dunet=1.9.1 \
  numpy scipy matplotlib pandas h5py opencv scikit-image \
  pyyaml tqdm tensorboard \
  -c pytorch -c nvidia -c conda-forge -y

# Activate environment
conda activate event_deflare

# Verify installation
python -c "from pytorch3dunet.unet3d.model import ResidualUNet3D; print('✅ E-DeflareNet ready')"

Quick Verification

# Test dataset generator
cd main
python main.py --test --num-sequences 2

# Test model inference (after training or downloading pretrained weights)
cd ../Unet_main
conda activate event_deflare
python inference_single.py --input <path_to_h5_file> --mode normal

Troubleshooting

CUDA version mismatch: Replace pytorch-cuda=12.1 with your CUDA version (e.g., 11.8)

pytorch-3dunet not found:

conda install -c conda-forge pytorch-3dunet=1.9.1

Import errors: Ensure you're in the correct directory and environment before running scripts.