# VR Mini Project 1 - Best Models Inference Payload

This repository contains the final, best-performing models and the standardized inference script (`predictor.py`) for the Visual Recognition Apparel Detection and Segmentation project.

## Models Included
Based on evaluation metrics, the following architectures were selected:
1. **Classification (Task 3.1):** `cls.pth` — Fine-tuned **ResNet-50**
2. **Detection & Segmentation (Task 3.2):** `seg.pt` — Fine-tuned **YOLOv8-seg**

## Categories Handled
Both models are configured to predict the following 5 canonical categories:
* `short sleeve top`
* `long sleeve top`
* `trousers`
* `shorts`
* `skirt`

## File Structure
```text
.
├── model_files/
│   ├── cls.pth       # ResNet-50 weights
│   └── seg.pt        # YOLOv8 weights
├── predictor.py      # Standardized inference wrapper
├── requirements.txt  # Dependencies (torch, torchvision, ultralytics, etc.)
└── validator_local.py# Local validation script for format checking
```

## Team Members
* **Dhruv Ramesh Joshi** – IMT2023032
* **Ankith Kini** – IMT2023075
* **Arnav Oruganty** – IMT2023078
* **Mithilesh Sai Yechuri** – IMT2023507