tear-cls

Setup

bash setup.sh

This activates .venv, runs uv sync --all-extras, downloads data.zip, unzips to data_raw/, then runs tearcls/data_split.py and tearcls/augmentation.py to produce data/processed/ with index.csv.

Reuse an existing data.zip instead of re-downloading:

bash setup.sh --skip-download

Loading data

from tearcls.data import build_loaders

train_dl, val_dl, test_dl = build_loaders(batch_size=16, num_workers=2)

for batch in train_dl:
    batch["image"], batch["label_idx"], batch["label"]

Train uses a class-balanced sampler and includes pre-dumped augmented variants; val/test are deterministic and originals-only. Pass load_augmented=False to train on originals only, or processed_dir=Path("data/processed_no_cmap") to switch ablation dumps.

Web UI & Demo Gallery

The project includes an interactive web UI that serves an artistic demo gallery of the AFM tear-film topography scans. It features:

• Interactive 3D surfaces: Rendered representations of the scans.
• Conversion to sound (sonification): Experimental sonification where deterministic acoustic parameters are extracted directly from the physical roughness of the tear-film topography.
• Live Classification: Demo integration calling the trained RF classifier.

To run the UI server locally:

uv run uvicorn tearcls.server:app
# or
python -m tearcls.server

Then, open http://127.0.0.1:8000 in your browser.