wassim249/PyTorch-Human-Segmentation

PyTorch Human Segmentation Project

This is a PyTorch-based project for human segmentation in images. Human segmentation is the task of identifying and segmenting humans in images, separating them from the background. This repository provides a template and guidelines for building your own human segmentation project using PyTorch.

Table of Contents

• Introduction
• Requirements
• Dataset
• Getting Started
  • Installation
  • Data Preparation
• Training
• Inference
• Evaluation
• Results
• Contributing
• License

Introduction

In this project, we use U-net architecture. The goal is to create a model that can accurately segment humans in images, which can have various applications such as object recognition, image editing, and more.

Requirements

You can install the required packages using the provided requirements.txt file.

pip install -r requirements.txt

Dataset

To train and evaluate your model, you will need to download the dataset from kaggle using this bash script: Dataset Credit : https://github.com/VikramShenoy97/Human-Segmentation-Dataset

bash ./dataset/download_dataset.sh

dataset/
    ├── Ground_Truth/
    │    ├── 1.jpg
    │    ├── 2.jpg
    │    └── ...
    └── Training_Images/
    |    ├── 1.png
    |    ├── 2.png
    |    └── ...
    |── train.csv

Getting Started

Installation

Clone this repository:

git clone https://github.com/wassim249/PyTorch-Human-Segmentation.git
cd PyTorch-Human-Segmentation

Install the required dependencies:

pip install -r requirements.txt

Data Preparation

Prepare your dataset as described in the Dataset section.

Training

Train your segmentation model using the following command:

```bash
python train.py \
  --img_size 128 \
  --data_dir path/to/dataset \
  --batch_size 32 \
  --shuffle True \
  --validation_split 0.2 \
  --num_workers 4 \
  --data_path path/to/dataset \
  --learning_rate 0.001 \
  --weight_decay 0.0001 \
  --ams_grad True \
  --epochs 50 \
  --save_dir checkpoints \
  --tensorboard True \
  --checkpoint_path None

You can customize the training parameters as needed.

Inference

Perform inference on new images using a pre-trained model:

python inference.py \
  --img_size 128 \
  --data_dir path/to/dataset \
  --batch_size 32 \
  --validation_split 0.2 \
  --num_workers 4 \
  --data_path path/to/dataset \
  --save_dir checkpoints \
  --checkpoint_path ./saved/checkpoint.pth

Evaluation

Evaluate your model's performance using appropriate metrics. You can provide guidance on how to evaluate the segmentation accuracy.

Results

Showcase the results of your human segmentation model, including example images with segmentation masks.

Issues

If you encounter any problems or have any inquiries about this project, please open an issue and I will try to respond as soon as possible.