Image Segmentation AI

## Image Segmentation AI

**Image segmentation is a core pixel-level analysis task in the field of computer vision**. By dividing the image into semantically homogeneous regions, it achieves precise separation of target contours and background. It covers sub-directions such as semantic, instance, and panoptic segmentation, providing underlying support for high-level visual tasks. As a key technology connecting image perception and decision-making, image segmentation breaks through the box-level limitations of target detection. With pixel-by-pixel annotation capabilities, it accurately characterizes target morphology and position, which is a necessary prerequisite for visual understanding in complex scenes.

## Module Principle
[Get mask or polygon point set, category, detection box for a certain area or defect by calling the deep learning library to infer the trained deep learning segmentation model]

### Algorithm Workflow
[First, build the deep learning model for segmentation, determine the data input form, training strategy, and model structure, and then determine the inference method and inference structure for this model. After the preparation work is done, collect the dataset according to the project requirements, then annotate the dataset, then perform training. After training, export the model, and then import the model into vg3 for operation.]

### Algorithm Formula

Convolution operation (downsampling, feature extraction)

Transposed convolution operation (upsampling, restoring size)

Loss function

Intersection over Union (IoU)

## Usage Scenarios
Defect detection

## Usage Method

### Pre/Post-Sequence Modules
Camera image acquisition, display image, etc.

### Usage Example
[Record operation video]

## Parameter Configuration Description
Add AI Model
  1. Click AI Model Tree to enter the AI model addition page.
![](/media/202601/2026-01-15_165502_1478870.0590558949650134.png)
  2. Click Add to open the Set AI Model page to add a model.
![](/media/202601/2026-01-15_165911_0731130.8435335254751559.png)
  3. Below, you can select algorithms from multiple companies (currently supported are Huitu Technology, Fengrui Lingchuang), and multiple types of models. For this image segmentation, select the segmentation category.
![](/media/202601/2026-01-15_165934_0218030.6019170157417704.png)

![](/media/202601/2026-01-15_165957_1083930.6417422314809366.png)
Function Input
  AI Model: Select the model chosen in the AI Model Tree.
  Minimum Threshold: Also called confidence threshold, it is a probability value for recognizing a certain target. When this probability value is lower than the set minimum threshold, the recognized target will be filtered out. Therefore, appropriately lower the minimum threshold when the target to be recognized is not recognized.
  Defect Drawing Mode: Includes draw defect when output is true, draw defect when output is false.
  Result Label Drawing: Includes do not draw recognition label, draw recognition label.
  Judgment Result Output: Includes judgment result output false when conditions are met, judgment result output true when conditions are met.
  Recognition Label Filter: When it is necessary to filter out a certain label, enter a certain label here to filter out results related to labels other than this label.
![](/media/202601/2026-01-15_170504_9706920.3898506192082045.png)

## Output Description
Function Output
![](/media/202601/2026-01-22_174412_6130760.6513917510157249.png)
## Function Attribute Description
Function Attributes
![](/media/202601/2026-01-22_174838_8506430.6111941053389512.png)
Basic Parameters

![](/media/202601/2026-01-22_175039_7869410.6778533495793185.png)
Run Parameters
![](/media/202601/2026-01-22_175112_6173550.07899279482637944.png)
## Other Supplements

## Precautions

---

Click here to jump to the corresponding Chinese manual page: [Image Segmentation AI](/doc/159)