● Covered by 1 source · 1 reportMedium impact1 neutral
MultiMatte Model Improves Image Background Removal with Alpha Mattes and Text Prompts
Key points
- MultiMatte uses alpha mattes for precise background removal.
- It allows text prompts to specify objects to keep or remove.
- The model improves on SAM 3's segmentation, scoring 0.901 S-measure.
- Fine-tuned 2.27% of SAM 3's parameters using LoRA.
Introduction of MultiMatte
MultiMatte is a new image background removal model that allows users to specify which objects to keep or remove using text prompts. This model is built upon SAM 3 (Meta, 2025) and focuses on improving the accuracy of image segmentation.
Enhanced Segmentation with Alpha Mattes
Traditional segmentation models, like SAM 3, use binary masks, classifying pixels as either inside or outside an object. This approach struggles with fine details or translucent elements. MultiMatte addresses this by employing alpha mattes, which assign a continuous opacity value to each pixel, allowing for more accurate representation of fuzzy boundaries.
On the DIS-VD benchmark, MultiMatte achieved a 0.901 S-measure, significantly outperforming SAM 3's 0.667. Across five high-resolution DIS splits, MultiMatte scored between 0.893 and 0.923, compared to SAM 3's scores of 0.649 to 0.703.
Training and Technical Details
MultiMatte was developed by fine-tuning 19.49 million of SAM 3's 860 million parameters, representing 2.27% of the total weights. This was achieved using low-rank fine-tuning with LoRA (Hu et al., 2021), which allowed the model to retain SAM 3's text alignment and vocabulary while improving segmentation.
The training involved 19,953 images covering diverse subjects like salient objects, camouflage, hair, and marine scenes. The model was trained for 14,000 steps using the semantic segmentation objective that SAM 3 already utilizes, which includes focal loss and Dice loss.
Impact on Image Editing and AI
The introduction of MultiMatte offers a more precise tool for image background removal, particularly beneficial for tasks involving complex object boundaries. Its ability to be guided by text prompts simplifies the process for users, potentially making advanced image editing more accessible. This advancement contributes to the ongoing development of more sophisticated and user-friendly AI-powered image manipulation tools.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
Primary sources
GitHub feyninc/nobgReporting from
A new model called MultiMatte has been introduced, which improves image background removal by using alpha mattes and allowing users to specify objects for removal with text prompts. This development enhances the precision of image segmentation, particularly for fine or translucent elements, which could benefit various applications requiring accurate object isolation.