MDSSD : Multi-scale Deconvolutional Single Shot Detector for Small Object

 

• To make the feature maps of the shallow layers contain more semantic information, we build several fusion layers between high-level and low-level feature maps through Fusion Blocks.
• conv11 and conv12 have totally lost the fine details of small objects, and therefore we apply the Fusion Block before conv11.
• In order to share the structure of Fusion Block, we delicately design symmetric topology between shallow layers and deep layers.
• In order to further improve the performance of small object detection, it is necessary to take full advantage of the shallow feature maps. Therefore, we add Fusion Module 1 which connects the lower layer conv3 3 and layer conv8 2 to make predictions.