Matting Segmentation in Xiaomi: a breakdown of technology with examples of customization

Matting segmentation is increasingly common in Xiaomi smartphone and smart appliance descriptions, but most users still don’t understand what it is and why. If you’ve ever used portrait mode, changed the background to a video, or tried the magic eraser feature in a gallery, you’ve seen the results of this technology. In this article, we’ll take a closer look at how matting segmentation works in Xiaomi devices, where it’s used in addition to photos, and why it’s one of the key technologies for modern gadgets.

Many people mistakenly think that matting is just a “background blur,” but it’s actually a complex process of pixelating objects in an image. Unlike primitive blurring, here the neural network analyzes each pixel, determining which object it belongs to (human, hair, clothing, background, etc.), and then allows these objects to be manipulated independently. Xiaomi has integrated this technology not only into smartphone cameras, but also into editing applications, smart displays and even some models of robot vacuum cleaners (for mapping real-world devices).

Let's just say: 2026 Xiaomi is using its own version of matting segmentation "AI Matting 2.0", who has been trained on a dataset with more than 10 The same version is available on smartphones with Snapdragon chips, and millions of annotated images, including complex cases (hair, translucent objects, complex lighting). 8 Gen 2/3 Dimensity 9000+, and also in some models of tablets Xiaomi Pad.

What is matting segmentation: simple words

The term matting segmentation consists of two parts: matting, the process of creating a transparency mask for an object, and segmentation, the separation of an image into logical parts, together a technology that allows objects to be distinguished in a photo or video with pixel-specific accuracy, preserving details like hair, fur or translucent edges.

By comparison, conventional segmentation (as in older versions of Portrait Mode) simply draws a rough mask around the person, while matting analyzes each pixel to determine its “belonging” to the object. For example, in a photo with loose hair, classical segmentation will cut them along the contour of the head, and matting will retain each strand, even if they merge with the background.

• 🔍 Accuracy: Matting works with resolution up to 4K It supports up to 16 levels of translucency (alpha channel).
• ⚡ Speed: Xiaomi 14 Ultra and Redmi smartphones K70 Pro processing takes less than 0.5 seconds per frame.
• 🤖 AI models: Hybrid neural networks are used (CNN + Transformer, trained on data from Leica and Xiaomi Imaging Lab.

The main difference from the competition: Xiaomi integrated matting not only into the camera, but also into system services. MIUI 15 is a function "AI Eraser, which uses this technology to remove objects from photos without traces.

Where is matting segmentation used in Xiaomi devices

If you think matting is just about blurring the background, you're missing 80 percent of the possibilities. In the Xiaomi ecosystem, the technology is being used in five key scenarios:

1. Bokeh, Night Portrait and Dual Video Modes in Xiaomi Smartphones 13T Pro and newer.
2. Photo/video editing: Mi Gallery Apps, CapCut (built-in version for Xiaomi) and AI Photo Editor.
3. Video Calls: Background Replacement in Mi Video Call and Integration with Zoom/Teams.
4. Smart displays: Face segmentation for personalized notifications on Xiaomi Smart Display.
5. Robots and IoT: Mapping rooms in Xiaomi Robot Vacuum-Mop 2 Ultra (recognition of furniture, carpets, obstacles).

The case of robot vacuum cleaners is particularly interesting: matting helps distinguish stationary objects (walls, furniture) from dynamic ones (people, pets), for example, if a cat runs past a vacuum cleaner, the system will not consider it an obstacle, but simply bypass without changing the map of the room.

💡

In smartphones Xiaomi with Snapdragon 8 Gen 3 matting works in real time even in the smartphone. 4K-To check for support, go to the camera settings. → Additionally. → AI modes and find a point "AI Matting".

How does matting segmentation work: technical details

Xiaomi's matting algorithm has three stages, and each of them is optimized for hardware specific devices.

Pre-processing (Pre-processing):

The frame is broken into blocks 256×256 For each of these pixels, the neural network predicts a depth map and semantic segmentation (where the face, body, background, etc.) in Xiaomi smartphones, this stage is accelerated by the Hexagon. DSP (in Snapdragon, or APU 690 (Dimensity).

2. Matting (actually highlight):

Here we use a hybrid model:

- CNN (convolutional neural network processes local parts (hair, edges).

Transformer analyzes the global context (e.g., understands that the hand belongs to a person, even if it is partially closed).

The result is an alpha matte with 16 levels of transparency.

3. Post-processing (Refinement):

Artifacts (such as "holes" in the hair) are removed, the mask is smoothed and applied to the original image. MIUI This step can be configured manually: for example, in Mi Gallery there is a slider "Edge Precision" (Edge Precision).

Interesting fact: in Xiaomi 14 Ultra matting is used for additional NPU-a chip (neural processor) that processes 20 TOPS (It allows for real-time effects even when shooting video in a video. 8K@24fps.

How to Enable and Configure Matting Segmentation on Xiaomi

Features based on matting in Xiaomi can be hidden in different menus depending on the model. Below is a universal guide for smartphones on the Internet. MIUI 14/15:

In the cell (portrait mode):

1. Open the Camera app.
2. Select Portrait Mode (Portrait).
3. Touch the badge. ⚙️ (setting) → turn on AI Matting or Leica Authentic Look (for Leica models).
4. For video: Go to Additional → Dual Video and activate Bokeh Effect.

2. In the gallery (edited photo):

Open photo → Press "Edit"

Select a tool "AI Eraser or Matting"

Contour the object for isolation or removal

Customize the accuracy of edges (Edge Precision slider)

Keep the result-->

In video calls (replacement of background):

1. When you call Mi Video Call or Zoom, click on three dots (⋮).
2. Select Background Effects (Background Effects).
3. Activate. AI Matting and select background (blurring, image or video).

⚠️ Note: On budget models (e.g. Redmi 12), matting works only for photos, and video requires a chip at least Snapdragon. 778G+ Or Dimensity 1080. Check for support in the Settings menu. → The phone. → Full specifications → Multimedia.

Comparison of matting in Xiaomi and competitors (Samsung, Apple, Google)

All top manufacturers have segmentation technologies, but the approaches are different.

Xiaomi’s main advantage is the openness of settings, for example, in Mi Gallery you can manually adjust the mask after auto-selection, whereas in iPhone or Pixel, the user can only access the finished results.

Another plus is cross-platform. Matting from Xiaomi works not only in native applications, but also in third-party ones (for example, in CapCut or in the Internet). VSCO), If they support Xiaomi Imaging SDK.

💡

Xiaomi uses a unique approach called Dynamic Matting, which adapts to the scene: for example, in portrait mode in low light, the algorithm automatically increases the weight of data from the ToF sensor (if it is in the smartphone).

Frequent problems with matting and how to solve them

Even the most advanced technology sometimes fails, and here are the typical mistakes and ways to fix them:

• 👻 Artifacts on edges: Problem: Hair or clothing look "ragged." Solution: In Mi Gallery, increase the Edge Smoothing setting or manually finish the mask with the Brush tool).
• 🌪️ Slow video processing: Problem: Applying effects takes a few seconds per frame.Solution: Lower video resolution to 1080p In camera settings or use Fast Matting mode with less accuracy.
• 🔴 Not working in third-party apps: Problem: There is no background replacement option on Zoom or Instagram.Solution: Install the latest version MIUI Check if the Xiaomi Virtual Background app supports API. List of compatible programs: Settings → Special facilities → Virtual background.

⚠️ Note: If after the update MIUI matting effects stopped working, reset the camera settings: Settings → Annexes → Camera. → Warehouse → Clean up the data. It'll bring back the factory parameters of the neural network.

For owners of older smartphones (for example, Xiaomi) 11T) There's a workaround: use the Focos app (available in GetApps), which emulates matting software, although it's slower.

The future of matting segmentation in Xiaomi: what to expect in 2026

According to internal sources Xiaomi (and presentations on the MWC 2026), in the next generation of devices matting will receive several key improvements:

• 🤖 3D-matting: Segmentation of objects in three-dimensional space for AR-applications (for example, virtual fitting of clothes with lighting and shadows in mind).
• 🎥 Real time in the 8K: Video processing without delay on Snapdragon 8 Gen 4 chips.
• 🧠 Personal models: The neural network will learn from your photos, improving accuracy for your face and frequently filmed objects.
• 🌍 Cross-platform: Integration with Windows and macOS via Xiaomi HyperOS (e.g., replacing the background in Zoom on a PC using smartphone matching).

In 2026, Xiaomi plans to release a separate API For developers, which will allow the use of matting in games and applications of virtual reality, a prototype is already being tested, where a neural network in real time separates the player from the background and embeds him in the background. 3D-world.

Also expected to appear matting in smart glasses (project Xiaomi Smart Glasses 2). AR-Objects with depth of scene.

FAQ: Answers to Frequent Questions about Matting Segmentation