Spheroid Validator (Valid vs. Invalid)

Binary image classifier that labels 3D spheroid images as valid or invalid using a fine-tuned ResNet-50.
Includes an inference script that scans a Test/ folder, predicts a class for each image, and writes results to CSV.

Overview

This repository provides a ResNet50 trained model to classify spheroid images into valid or invalid.
Images are preprocessed and passed through a fine-tuned ResNet-50.
Predictions are saved to predictions.csv with class label and confidence.

Features

• ResNet-50 backbone with a custom classification head (dropout + linear).
• Batch inference over a folder (./Test/) of images.
• Outputs a CSV with filename, predicted_class, confidence.
• Flexible class naming:
  • Auto-detect from Dataset/train/<class_name>/... (ImageFolder layout), or
  • Provide explicitly via --classes "invalid,valid".
• Supports common image formats: .jpg, .jpeg, .png, .bmp, .tif, .tiff.

Setup and Installation

1. Python 3.9–3.12 recommended.
2. Install dependencies:

pip install torch torchvision pillow