Anchor-Free Segmentation Model for Partially Supervised Mask Generalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional segmentation models face challenges in achieving strong mask generalization when trained with partially supervised data, where some classes have complete ground truth data while others have less informational data, leading to reduced performance on unseen classes.
Innovation Solution
A machine-learned segmentation model incorporating an anchor-free detector model and a deep mask head network with an encoder-decoder structure, including multiple layers, which enables improved generalization without additional losses or specialized modules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If partially supervised training data is used, then data collection cost is reduced, but segmentation performance on unseen classes deteriorates
Solution Approach 1:
The model segments the training data into two distinct types: fully-labeled data for seen classes and partially-labeled data for unseen classes. This segmentation allows the model to learn from both high-quality and low-quality data without performance degradation on unseen classes, resolving the contradiction between data collection cost and segmentation performance.
Solution Approach 2:
The patent introduces a mask quality parameter that varies across different training samples. By changing this parameter to reflect the quality of ground truth labels (complete vs. partial), the model learns to adapt its processing strategy based on data quality, thereby maintaining high performance on unseen classes while using cost-effective partial labeling.
2Reliability
If fully supervised training is used, then segmentation performance is improved, but data collection cost increases
Solution Approach 1:
Instead of requiring complete ground truth labels for all classes (fully supervised approach), the patent applies partial labeling where only seen classes receive complete labels and unseen classes receive partial labels. This partial action approach reduces data collection cost while maintaining acceptable performance, especially for unseen classes.
3Measurement precision
If anchor-based detector models are used, then detection accuracy is improved, but model complexity increases
Solution Approach 1:
The patent extracts and removes the complex anchor-based components from the detector model. By taking out the anchor mechanism and using a simpler anchor-free approach with a deep mask head network, the model achieves comparable detection accuracy with reduced complexity, resolving the contradiction between measurement precision and device complexity.
Data Source
AI summary
A computer-implemented method for partially supervised image segmentation having improved strong mask generalization includes obtaining, by a computing system including one or more computing devices, a machine-learned segmentation model, the machine-learned segmentation model including an anchor-free detector model and a deep mask head network, the deep mask head network including an encoder-decoder structure having a plurality of layers. The computer-implemented method includes obtaining, by the computing system, input data including tensor data. The computer-implemented method includes providing, by the computing system, the input data as input to the machine-learned segmentation model. The computer-implemented method includes receiving, by the computing system, output data from the machine-learned segmentation model, the output data including a segmentation of the tensor data, the segmentation including one or more instance masks.