Geometrically regularized automatic differentiation framework for BCDI phase retrieval.

• Python ≥ 3.12
• NVIDIA GPU with CUDA ≥ 12 and compatible drivers

Note: tensorflow[and-cuda] and tensorrt require NVIDIA hardware. Installation on CPU-only machines or non-NVIDIA GPUs is not supported.

python3.12 -m venv .venv
source .venv/bin/activate   # on Windows: .venv\Scripts\activate

pip install git+https://github.com/matteomasto/convex-ad.git

This will install convex_ad and all its dependencies, including TensorFlow with CUDA support, and TensorRT.

If you want to develop or modify the code:

git clone https://github.com/matteomasto/convex-ad.git
cd convex-ad
pip install -e .

import convex_ad
print(convex_ad.__version__)   # should print 0.1.0

See the examples/ folder and the notebooks/ folder for a full demo.

import numpy as np
import convex_ad
from convex_ad.core import PhaseRetrievalModel, train_step
from convex_ad.losses import total_loss

# Load your diffraction data
Iobs = np.load("data.npz")["I"].astype(np.float32)

# Build and train the phase retrieval model
model = PhaseRetrievalModel(Iobs, batch_size=4)
# ... see examples/run_reconstruction.py for a complete workflow

See LICENSE.