Francesco2035/fcos-pneumonia-detection-model-checkpoints

FCOS Pneumonia Detection — Model Checkpoints

This repository contains the pretrained checkpoints of four FCOS-style pneumonia detection models developed for the Numerical Analysis for Machine Learning project at Politecnico di Milano.

Models

Four detector configurations are provided:

• R50-IN: ResNet-50 + FCOS-style detector, initialized with ImageNet pretrained weights
• R50-CX: ResNet-50 + FCOS-style detector, initialized with Chest-Xray pretrained weights
• R101-IN: ResNet-101 + FCOS-style detector, initialized with ImageNet pretrained weights
• R101-CX: ResNet-101 + FCOS-style detector, initialized with Chest-Xray pretrained weights

All models are trained for pneumonia detection and localization on the RSNA Pneumonia Detection Challenge dataset.

Model Configuration

Model	Backbone	Initialization	Task
R50-IN	ResNet-50	ImageNet	Pneumonia detection and localization
R50-CX	ResNet-50	Chest-Xray	Pneumonia detection and localization
R101-IN	ResNet-101	ImageNet	Pneumonia detection and localization
R101-CX	ResNet-101	Chest-Xray	Pneumonia detection and localization

Common configuration:

• Detector: FCOS-style anchor-free object detector
• Input resolution: 512 × 512
• Dataset: RSNA Pneumonia Detection Challenge

Usage

The checkpoints are intended to be used with the source code available in the corresponding GitHub repository.

For example, to analyze an R101-IN checkpoint:

uv run python main.py \
    --mode analyze \
    --checkpoint path/to/resnet101_imagenet_long_ft.pt \
    --backbone imagenet \
    --resnet-depth 101