Skip to content

Google Colab Testing & Evaluation

For automated large-scale evaluation of exported models on Google Colab, the project provides a dedicated testing notebook.

Notebook location: colab/cattle_buffalo_tester.ipynb

Features

This notebook allows you to:

  1. Test Exported Models: Load your exported ONNX model (lite2_fp32.onnx).
  2. Single & Batch Inference: Run inference on single images or entire zipped batches.
  3. HTML Reports: Generate comprehensive HTML evaluation reports containing confusion matrices, per-breed accuracy tables, and misclassified examples.
  4. Large-Scale Kaggle Evaluation: Automatically download the complete Kaggle dataset (algsoch/breed-cattle-buffalo) directly to Colab and run inference on all images, yielding full dataset metrics and per-breed accuracy tables.

Usage

Option 1: Uploading local test split

The repository includes a helper script create_test_eval_zip.py that you can run locally to bundle your generated test split images into a zip file (test_eval_images.zip). Upload this file to your Colab session for batch evaluation.

Option 2: Full Dataset Evaluation via Kaggle

Using the §9 — Large-Scale Kaggle Dataset Evaluation section in the notebook, you can supply your Kaggle API credentials. The notebook will download the entire dataset and map breeds to your model's class maps automatically, providing a full-scale performance report.