Amodal Mask Segmentation via Temporal Feature Propagation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing segmentation methods struggle to accurately segment objects that are partially occluded, as they fail to effectively incorporate temporal information and propagate features from visible areas to occluded areas.
Innovation Solution
A segmentation method that receives image frames, determines a feature map based on temporal information, extracts features of a region of interest (ROI), predicts the class of an object, and segments instances by correcting an amodal mask using a combination of modal and occluded masks weighted by confidence maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing segmentation methods are used, then processing is simpler, but accuracy in segmenting occluded objects deteriorates
Solution Approach 1:
The patent segments the segmentation task into multiple components: extracting features from visible areas, predicting amodal masks for occluded areas, generating confidence maps, and combining results through weighting. This segmentation of the processing task enables accurate handling of occluded objects while maintaining manageable system complexity through modular design.
Solution Approach 2:
The patent performs preliminary actions by extracting features from visible areas and predicting amodal masks before final segmentation. The confidence maps are generated in advance to guide the combination process. These preliminary computations enable accurate segmentation of occluded objects by preparing necessary information beforehand.
2Measurement precision
If temporal information is incorporated, then segmentation accuracy improves, but processing time increases
Solution Approach 1:
The patent applies partial action by selectively incorporating temporal information only where needed for occluded area prediction, rather than processing all frames uniformly. The system uses adjacent frames to generate confidence maps and amodal masks only for regions requiring occlusion handling, reducing overall processing time while maintaining accuracy improvements from temporal context.
3Measurement precision
If features are propagated from visible to occluded areas, then amodal mask prediction improves, but computational complexity increases
Solution Approach 1:
The patent introduces confidence maps as an intermediary that mediates between the propagated features and the final segmentation result. The confidence maps weight the contribution of propagated features from visible areas to the final amodal mask prediction, enabling accurate feature propagation while controlling computational complexity through selective application.
Data Source
AI summary
A segmentation method and segmentation apparatus are provided, where the segmentation method includes receiving image frames comprising a current frame and an adjacent frame to the current frame, determining a feature map to aggregate the image frames based on temporal information between the current frame and the adjacent frame, extracting a feature of a region of interest (ROI) corresponding to instances included in the current frame from the feature map, predicting a class of an object corresponding to the ROI based on the feature of the ROI, and segmenting the instances by correcting an amodal mask predicted corresponding to the class of the object based on the feature of the ROI.