Pixel-Level Object Detection Using Segmented Area and Detail Units
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pixel-level object detection methods face a decrease in inference accuracy due to high annotation costs and the need for detailed training data, which is not effectively addressed by prior methods that rely heavily on accurate initial pixel selection and cannot maintain high accuracy while reducing annotation costs.
Innovation Solution
A pixel-level object detection system that integrates local and area detection units to produce a segmentation map, using a detail detection unit for local information and an area detection unit based on learning data, with a result integration processing unit to combine outputs and reduce the reliance on detailed training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If detailed pixel-level training data is used, then inference accuracy is improved, but annotation cost increases
Solution Approach 1:
The detection process is segmented into two independent stages: area detection (coarse-level) and detail detection (fine-level). Each stage processes different aspects of the image independently, allowing the system to achieve high inference accuracy without requiring complete pixel-level annotations for training, thereby reducing annotation costs while maintaining measurement precision.
Solution Approach 2:
The system transitions from a single-stage pixel-level detection approach to a two-stage hierarchical approach that operates at different levels of detail. By adding the dimension of multi-scale processing (area level + detail level), the system achieves accurate pixel-level detection without requiring proportionally expensive pixel-level training annotations.
2Productivity
If pixel groups are formed in step one, then processing efficiency is improved, but pixel groups containing mixed target and non-target pixels cannot be refined in step two
Solution Approach 1:
The detection process is segmented into two independent stages: area detection (coarse-level) and detail detection (fine-level). Each stage processes different aspects of the image independently, allowing the system to achieve high inference accuracy without requiring complete pixel-level annotations for training, thereby reducing annotation costs while maintaining measurement precision.
Solution Approach 2:
The detail detection unit applies specialized processing focused specifically on pixels within detected areas, using local information from the inference image to refine classifications. This localized approach allows refinement of pixel groups that were initially mixed, improving measurement precision without compromising the overall processing efficiency gained from the initial area detection stage.
Data Source
AI summary
A pixel-level object detection system that detects an object at a pixel level, includes: an imaging unit that acquires an inference image that is an image that captures a detection target; an area detection unit that detects an area including a detection target from the inference image; a detail detection unit that detects the detection target using only local information from the inference image; and a result integration processing unit that integrates an output of the area detection unit with an output of the detail detection unit to output a segmentation map indicating, as a probability map, which pixel in the image corresponds to the detection object.