Trains a conditional GAN on SVG gem-cutting diagrams and metadata, then generates new diagram images from metadata inputs.

Gem-cutting diagram generation is a niche problem with no off-the-shelf dataset or model. GemDiagramAI provides the full ML pipeline — data preparation, cGAN training on 512x512 images, and inference — so the workflow can be iterated on with real lapidary diagram data. The project is designed for Mac M1/M2 (uses tensorflow-macos and tensorflow-metal).

# Set up a virtual environment (required on Apple Silicon)
pip install virtualenv
virtualenv tf_m1_env
source tf_m1_env/bin/activate
pip install -r requirements.txt

# 1. Prepare training data from SVG images + JSON metadata
python prepare_data_for_training.py
# Reads SVGs and a JSON metadata file; writes .npz files to training-data/

# 2. Train the model
python train_model.py
# Saves generator checkpoints every 20 epochs and a final generator_model_final.h5

# 3. Prepare generation inputs
python prepare_data_for_generation.py --save_dir generation-data

# 4. Generate new diagrams
python run_model.py --generation_data_dir generation-data
# Prompts for the path to a trained generator model (e.g. generator_model_final.h5)

• model.py — defines the cGAN architecture. The generator upsamples a 100-dimensional noise vector through two UpSampling2D + Conv2D blocks to produce 512x512 RGB images. The discriminator is a four-layer strided-convolution network with LeakyReLU and dropout. Both use Adam (lr=0.0002, beta=0.5).
• train_model.py — orchestrates the adversarial training loop for 2000 epochs (configurable in constants.py), saving generator checkpoints every 20 epochs.
• data_utils.py — preprocessing utilities for loading and normalizing SVG-derived image data.
• prepare_data_for_training.py / prepare_data_for_generation.py — convert raw SVGs + metadata JSON into .npz arrays consumed by the training and inference scripts.
• run_model.py — loads a saved generator and produces diagram images from metadata-conditioned inputs.
• constants.py — central config: IMAGE_SIZE=512, EPOCHS=2000, BATCH_SIZE=32, SAVE_INTERVAL=20.
• Training logs to TensorBoard (./logs/).

Edit constants.py to change image resolution, epoch count, batch size, or checkpoint frequency. All paths default to training-data/ and generation-data/ but can be overridden with CLI arguments.

Experimental. The model architecture and training pipeline are implemented, but results depend on the quality and quantity of SVG training data supplied by the user. No pre-trained weights are included in the repository.

MIT