ZHKKKe/MODNet
MODNet
MODNet is a deep learning image segmenter and portrait matting model designed to isolate human subjects from backgrounds. It generates high-quality alpha mattes for images and video using only standard RGB input, removing the requirement for manual trimap guides.
The framework is optimized for real-time inference and provides utilities to export pre-trained model weights into specialized formats for deployment on target hardware.
The project covers the full workflow for portrait isolation, including supervised matting model training on labeled datasets, real-time video background removal, and edge hardware deployment.
Features
- Portrait Segmentation - Isolates human portraits from backgrounds using a specialized deep learning segmentation model.
- AI Subject Isolation - Extracts human silhouettes by identifying the primary subject using an RGB-based segmentation model.
- Trimap-Free Alpha Estimation - Predicts high-quality alpha transparency masks directly from RGB input without requiring manual trimap guides.
- Neural Networks and Deep Learning - Implements a deep neural network architecture for pixel-level subject isolation and alpha matte generation.
- Deep Learning Video Matting Tools - Implements a deep learning framework to separate human subjects from backgrounds in images and video.
- Image Background Removal - Generates alpha mattes to isolate foreground subjects from backgrounds using only standard RGB input.
- Real-Time Portrait Matting - Removes backgrounds from images and videos of people in real time without requiring manual trimaps.
- Temporal Video Background Removal - Isolates human subjects from video streams while ensuring temporal consistency across frames for clean alpha mattes.
- Detection Model Training - Trains custom portrait matting and segmentation models using supervised training on labeled datasets.
- Inference Model Deployment - Converts pre-trained matting weights into optimized compute graphs for efficient hardware execution.
- Edge Hardware Optimizations - Optimizes model weights and formats to reduce inference latency and memory footprint on resource-constrained edge devices.
- Matting Model Training - Supports training and optimizing portrait matting models using supervised datasets of labeled image pairs.
- On-Device Inference Exports - Provides utilities to export pre-trained model weights into formats optimized for edge hardware deployment.
- Visual Stream Processing - Analyzes video frames sequentially with low latency to maintain consistent subject isolation during playback.
- Supervised Model Weight Optimization - Optimizes model weights for accurate edge detection using supervised training on labeled image pairs.
- Multi-Format Exporters - Provides a toolkit to convert models into multiple optimized formats for cross-platform hardware deployment.
- Background removal - Listed in the “Background removal” section of the Ailia Models awesome list.
