Multilabel-Classification_CNN_and_Vision-transform

📌 Overview

This repository implements an Image Multi-Label Classification model, which can identify multiple labels in a single image. Unlike traditional image classification, where an image is assigned a single category, multi-label classification allows an image to have multiple labels simultaneously.

🔍 Difference Between Image Classification and Multi-Label Classification

🚀 Features

• Uses a deep learning model (CNN, ResNet,VIT etc.) for multi-label classification.
• Implements Binary Cross-Entropy (BCE) Loss to handle multiple labels per image.
• Data preprocessing and augmentation for better generalization.
• Evaluation metrics: F1-score, Precision, Recall, mAP (mean Average Precision).
• Supports custom datasets and pretrained models for fine-tuning.

Dataset Link

The dataset used for training contains 16170 images of 8 different clothing categories in 9 different colours.

This repository contains resources for training a deep learning model to multi-label classification. It includes two main Jupyter notebooks for model training, each implementing a distinct architecture:

1. Custom Architecture Notebook : This notebook demonstrates the use of a custom-built neural network architecture tailored specifically for multi-label classification. Designed for flexibility and simplicity, the custom architecture allows for experimentation and adaptation to varying datasets.

   Model inference:

3. VIT Architecture Notebook : In this notebook, I'll show how one can fine-tune any pre-trained vision model from the Transformers library for multi-label image classification.

   refference

   Model inference: