README.md

---
license: apache-2.0
tags:
- segmentation
size_categories:
- 10K<n<100K
---
# CHIP Dataset - Segmentation Training Guide

## Overview
This dataset is designed for training segmentation models, specifically for human/CHIP segmentation tasks using YOLO models.

## Dataset Structure
After extracting `CHIP_dataset.zip`, the directory structure should include:
- `train/` and `valid/` folders with images
- `data.yaml` configuration file for YOLO training

## Prerequisites
- Python environment
- GPU with CUDA support (recommended)

## Installation
```bash
pip install ultralytics tqdm
```

## Download and Extract Dataset
```bash
# Download dataset
wget https://huggingface.co/datasets/OSAS-AI/CHIP-Segmentation/resolve/main/CHIP_dataset.zip

# Extract
from zipfile import ZipFile
from tqdm import tqdm
import os

zip_path = "CHIP_dataset.zip"
extract_dir = "."
os.makedirs(extract_dir, exist_ok=True)

with ZipFile(zip_path, "r") as zip_ref:
    members = zip_ref.infolist()
    for member in tqdm(members, desc="Extracting", unit="file"):
        zip_ref.extract(member, extract_dir)

print(f"Extracted to: {extract_dir}")
```

## Download Pre-trained Model
```bash
wget https://huggingface.co/Ultralytics/YOLO26/resolve/main/yolo26s-seg.pt
```

## Training
```python
from ultralytics import YOLO

# Load model
model = YOLO('yolo26s-seg.pt')  # small variant – good balance

# Train
results = model.train(
    data='data.yaml',
    epochs=300,
    imgsz=640,
    batch=32,  # reduce if OOM
    device="0,1",  # adjust for your GPUs
    workers=8,
    project='runs/train',
    name='human_segmentation',
    exist_ok=False,
    patience=15,
    save=True,
    val=True,
    plots=True
)
```

## Key Training Parameters
- **Model**: YOLOv26s-seg (segmentation)
- **Image Size**: 640x640
- **Epochs**: 300
- **Batch Size**: 32 (adjust based on VRAM)
- **Multi-GPU**: Supports `device="0,1"`

## Inference / Usage
After training, the best model will be saved in `runs/train/human_segmentation/weights/best.pt`

```python
model = YOLO('path/to/best.pt')
results = model('path/to/image.jpg')
```

## Dataset on Hugging Face
[CHIP-Segmentation](https://huggingface.co/datasets/OSAS-AI/CHIP-Segmentation)

For more details, refer to [Ultralytics YOLO documentation](https://docs.ultralytics.com/).


You can copy the entire block above and save it as `README.md`. Let me know if you want any changes!