Training New Models

Leona's Sketches supports training custom models, allowing users to tailor the framework to their unique datasets and artistic goals.

Dataset Preparation:

Organize datasets into the following structure:

• Images: Store sketches in .png or .jpg format.

• Annotations: Include labels or metadata in .json files for each image, specifying categories or styles.

Training Workflow:

1. Start training with: python train.py --data "dataset_path" --epochs 10

   1. Monitor progress through logs saved in the logs/ directory.

   2. Checkpoints are generated periodically for model evaluation and recovery.

   Fine-Tuning Tips:

   • Adjust learning rates and batch sizes in the configuration files.

   • Use pre-trained weights as a starting point for quicker convergence.