📄️ 1. Split Training Data
The split command organizes your annotated particle data into training and validation sets, preparing it for PartiNet model training. This step can either convert STAR files from manual picking sessions directly to YOLO format, or split existing YOLO labels into organized train/val directories.
📄️ 2. Train Dual Detectors
PartiNet's architecture requires a two-step training regime. The first step is to train PartiNet's dual detectors to identify particles in a micrograph.
📄️ 3. Train Adaptive Router
PartiNet's architecture requires a two-step training regime. This step assumes you have already trained PartiNet to identify particles. Next you will train PartiNet's adaptive router to discriminate between hard and easy micrographs for real-time routing during detection. This step is almost identical to the previous training step, except that your model weight MUST be a .pt file from step 1.
📄️ Training Output Reference
This page describes the output generated during PartiNet training (both Step 1 and Step 2).