Maison-Roast / Basil — Menu Item Classifier

This repository hosts the vision model that powers Basil, the menu classifier for Maison-Roast, a vintage-style coffee shop e-commerce app.

Non-commercial / research use only. See License section below — this restriction comes directly from the underlying training data, not from the model architecture.

Model

Architecture: EfficientNetV2-Small
Framework: TensorFlow / Keras
Task: Image classification, 41 classes (40 menu classes + 1 "unknown")
Formats provided:
- basil-menu-classifier.keras — native Keras format. Use for fine-tuning or retraining.
- basil-menu-classifier.h5 — legacy HDF5 checkpoint, for compatibility with older Keras/TF loaders.

Training data

Built via a Food-101 ETL pipeline (train/val/test manifests + label map) from the kmader/food41 Kaggle re-upload of the original Food-101 dataset (Bossard, Guillaumin, Van Gool, ECCV 2014, ETH Zürich).

Food-101's own license states its images come from Foodspotting and are not ETHZ's property: any use beyond scientific fair use must be negotiated with the original photo owners under Foodspotting's terms. Because that negotiation isn't practically possible at this scale, this model is released for research and educational use only — not for unrestricted commercial redistribution.

Citation: ```bibtex @inproceedings{bossard14, title = {Food-101 -- Mining Discriminative Components with Random Forests}, author = {Bossard, Lukas and Guillaumin, Matthieu and Van Gool, Luc}, booktitle = {European Conference on Computer Vision (ECCV)}, year = {2014} } ```

Usage

```python import tensorflow as tf

model = tf.keras.models.load_model("basil-menu-classifier.keras") predictions = model.predict(input_batch) ```

License

Research/educational use only. The model weights are shared openly for learning, portfolio review, and non-commercial experimentation. Commercial deployment on menu images beyond fair use would require resolving image rights with the original Foodspotting photo owners, per the Food-101 license terms.

Downloads last month: -