Class-Agnostic Object Segmentation Neural Network for Unknown Parts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional digital image editing systems are inefficient and inflexible in segmenting objects, particularly failing to accurately recognize unknown objects and object parts without prior semantic classification, and require extensive training and resource-intensive processes.
Innovation Solution
A class-agnostic object segmentation system that utilizes a neural network to segment objects and object parts in digital images without classifying them, generating object masks for all objects irrespective of semantic meaning, and automatically selecting objects or partial objects based on user requests.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional systems use prior known semantic classification to segment objects, then segmentation accuracy for known objects is improved, but the system fails to segment unknown objects or object parts and requires extensive training
Solution Approach 1:
The patent extracts the semantic classification requirement from the object segmentation process. Instead of requiring objects to be classified into predefined categories, the system directly segments objects based on their visual features, removing the constraint that limited conventional systems to only known object types
Solution Approach 2:
The neural network is designed to perform universal object segmentation across all object types without requiring separate models or training for different categories. A single class-agnostic model handles both known and unknown objects, as well as complete objects and object parts, making the system multi-functional
2Ease of manufacture
If conventional systems segment only objects with prior known semantic meaning, then training requirements are reduced, but the system becomes inflexible and cannot handle unknown objects or object parts
Solution Approach 1:
The system performs self-service by automatically adapting to segment any object type encountered in the input image without requiring external classification information or retraining. The model independently identifies and segments objects based on their visual characteristics alone
3Device complexity
If conventional systems require prior knowledge of object class to segment objects, then system complexity is reduced, but object selection efficiency and user interaction speed decrease
Solution Approach 1:
The patent segments the object recognition process into direct visual feature analysis without the intermediate step of semantic classification. This segmentation of the processing pipeline eliminates the bottleneck of requiring prior class knowledge while maintaining relatively simple system architecture
Data Source
AI summary
The present disclosure relates to a class-agnostic object segmentation system that automatically detects, segments, and selects objects within digital images irrespective of object semantic classifications. For example, the object segmentation system utilizes a class-agnostic object segmentation neural network to segment each pixel in a digital image into an object mask. Further, in response to detecting a selection request of a target object, the object segmentation system utilizes a corresponding object mask to automatically select the target object within the digital image. In some implementations, the object segmentation system utilizes a class-agnostic object segmentation neural network to detect and automatically select a partial object in the digital image in response to a target object selection request.