metadata
license: mit
library_name: pytorch
tags:
- background-removal
- image-segmentation
- computer-vision
- pytorch
- foreground-extraction
pipeline_tag: image-segmentation
Background Remover (BEN2 Base)
BEN2 Base is a deep learning model for automatic background removal from images.
The model predicts a foreground segmentation mask that can be used to remove or replace the background.
This repository contains the pretrained weights:
BEN2_Base.pth
The model can be used in:
- photo editing tools
- product image processing
- portrait segmentation
- dataset preprocessing
- AI image pipelines
Model Details
| Property | Value |
|---|---|
| Model Name | BEN2 Base |
| Task | Background Removal |
| Architecture | Segmentation Network |
| Framework | PyTorch |
| File Size | 1.13 GB |
| Input | RGB image |
| Output | Foreground mask |
Repository Files
| File | Description |
|---|---|
| BEN2_Base.pth | Pretrained background removal model weights |
Installation
Install required libraries:
pip install torch torchvision pillow numpy opencv-python
Usage Example
Example inference using PyTorch.
import torch
from PIL import Image
import torchvision.transforms as transforms
# Load model
model = torch.load("BEN2_Base.pth", map_location="cpu")
model.eval()
# Preprocessing
transform = transforms.Compose([
transforms.Resize((512, 512)),
transforms.ToTensor()
])
image = Image.open("input.jpg").convert("RGB")
input_tensor = transform(image).unsqueeze(0)
# Inference
with torch.no_grad():
output = model(input_tensor)
mask = output.squeeze().cpu().numpy()
You can apply the mask to generate a transparent PNG or replace the background.
Example Workflow
- Load an image
- Resize and normalize
- Run model inference
- Generate segmentation mask
- Remove background
Use Cases
E-commerce
Remove backgrounds from product images.
Portrait Editing
Create clean profile images.
Content Creation
Prepare images for thumbnails, ads, or designs.
AI Pipelines
Preprocess images for ML datasets.
Limitations
- Performance may vary with extremely complex backgrounds.
- Very small foreground objects may reduce segmentation quality.
- Images should be resized for optimal results.
Training
This repository provides pretrained weights only.
License
Please verify the license before using the model in commercial applications.
Author
Ashank Gupta