therutvikpanchal/rmbg-img

BRIA Background Removal v2.0 Model Card

RMBG v2.0 is our new state-of-the-art background removal model significantly improves RMBG v1.4. The model is designed to effectively separate foreground from background in a range of categories and image types. This model has been trained on a carefully selected dataset, which includes: general stock images, e-commerce, gaming, and advertising content, making it suitable for commercial use cases powering enterprise content creation at scale. The accuracy, efficiency, and versatility currently rival leading source-available models. It is ideal where content safety, legally licensed datasets, and bias mitigation are paramount.

Developed by BRIA AI, RMBG v2.0 is available as a source-available model for non-commercial use.

Model Details

#####

Model Description

• —Developed by: BRIA AI
• —Model type: Background Removal

• —Model Description: BRIA RMBG-2.0 is a dichotomous image segmentation model trained exclusively on a professional-grade dataset.
• —BRIA: Resources for more information: BRIA AI

Training data

Bria-RMBG model was trained with over 15,000 high-quality, high-resolution, manually labeled (pixel-wise accuracy), fully licensed images. Our benchmark included balanced gender, balanced ethnicity, and people with different types of disabilities. For clarity, we provide our data distribution according to different categories, demonstrating our model’s versatility.

Distribution of images:

Qualitative Evaluation

Open source models comparison [image] [image]

Architecture

RMBG-2.0 is developed on the BiRefNet architecture enhanced with our proprietary dataset and training scheme. This training data significantly improves the model’s accuracy and effectiveness for background-removal task.<br> If you use this model in your research, please cite:

@article{BiRefNet,
  title={Bilateral Reference for High-Resolution Dichotomous Image Segmentation},
  author={Zheng, Peng and Gao, Dehong and Fan, Deng-Ping and Liu, Li and Laaksonen, Jorma and Ouyang, Wanli and Sebe, Nicu},
  journal={CAAI Artificial Intelligence Research},
  year={2024}
}

Requirements

torch
torchvision
pillow
kornia
transformers

Usage

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->

from PIL import Image
import matplotlib.pyplot as plt
import torch
from torchvision import transforms
from transformers import AutoModelForImageSegmentation

model = AutoModelForImageSegmentation.from_pretrained('briaai/RMBG-2.0', trust_remote_code=True)
torch.set_float32_matmul_precision(['high', 'highest'][0])
model.to('cuda')
model.eval()

# Data settings
image_size = (1024, 1024)
transform_image = transforms.Compose([
    transforms.Resize(image_size),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])

image = Image.open(input_image_path)
input_images = transform_image(image).unsqueeze(0).to('cuda')

# Prediction
with torch.no_grad():
    preds = model(input_images)[-1].sigmoid().cpu()
pred = preds[0].squeeze()
pred_pil = transforms.ToPILImage()(pred)
mask = pred_pil.resize(image.size)
image.putalpha(mask)

image.save("no_bg_image.png")