Object Detection Model Adaptive Penalty Co-occurrence Bias
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Residential camera systems face accuracy issues due to co-occurrence bias, where objects in images, such as humans and animals, are misidentified due to their frequent appearance together, leading to false alarms and reduced detection accuracy.
Innovation Solution
An adaptive penalty scheme is introduced into the object detection model's training process, using a scaling factor based on the distance between detected and ground truth bounding boxes to penalize false detections caused by co-occurrence bias, thereby improving the model's accuracy by reducing false alarms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If an object detection model is trained on a large dataset of images with co-occurring objects, then the model can detect objects using context information, but the model generates false alarms due to co-occurrence bias
Solution Approach 1:
The patent applies preliminary anti-action by introducing a penalty mechanism during training that proactively counteracts co-occurrence bias before it causes false alarms. The penalty is calculated based on the distance between detected objects and their ground truth positions, and is applied to the loss function to prevent the model from learning incorrect co-occurrence patterns. This anticipatory correction resolves the contradiction by eliminating the harmful effect before it manifests in deployment.
2Measurement precision
If co-occurrence patterns are used to improve object detection through context information, then detection accuracy improves, but precision decreases due to false positives near true detections
Solution Approach 1:
The patent applies local quality by making the penalty application spatially dependent. The penalty is not uniformly applied to all detections but is specifically targeted at false positives that occur near true detections based on their distance relationship. This localized correction allows the model to maintain context-based detection accuracy while precisely eliminating false positives in specific spatial regions, thus resolving the contradiction between detection accuracy and detection precision.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for model co-occurrence object detection. One of the methods includes accessing, for a training image, first data that indicates a detected bounding box for a first object depicted in the training image and a predicted type label, accessing, for the training image, ground truth data for one or more ground truth objects, determining, using the first data and the ground truth data, that i) the detected bounding box represents an object that is not a ground truth object represented by the ground truth data or ii) the predicted type label for the first object does not match a ground truth label for the first object identified by the ground truth data, determining a penalty to adjust the model using a distance between the detected bounding box and the labeled bounding box, and training the model using the penalty.