Enhancement and Fusion of Multi-Scale Feature Maps for Small Object Detection
Zhijun Xue, Wenjie Chen, Jing Li
Abstract
In recent years, deep convolutional neural networks have made breakthrough progress in object recognition and object detection tasks in the field of computer vision, and have achieved great results both in accuracy and speed. However, the detection of small objects is still difficult in the field of object detection, and the accuracy on the common dataset MS COCO is very low. This paper briefly reviews some work in multi-scale object detection algorithms, and then proposes a method of feature enhancement and fusion based on multi-scale feature maps, improving detection accuracy of small objects on MS COCO.
Topics & Concepts
Object detectionComputer scienceArtificial intelligenceConvolutional neural networkObject (grammar)Feature (linguistics)Pattern recognition (psychology)Scale (ratio)Feature extractionField (mathematics)Computer visionObject-class detectionViola–Jones object detection frameworkFusionFace detectionMathematicsGeographyCartographyPhilosophyLinguisticsFacial recognition systemPure mathematicsAdvanced Neural Network ApplicationsVideo Surveillance and Tracking MethodsAdvanced Image and Video Retrieval Techniques