Source-linked AI summary
MDSSD: Multi-scale Deconvolutional Single Shot Detector for Small Objects
Lisha Cui, Rui Ma, Pei Lv, Xiaoheng Jiang, Zhimin Gao, Bing Zhou, Mingliang Xu
TL;DR
Small-object detection remains challenging because top-level features retain little information about small objects, and upsampling has difficulty restoring lost spatial detail. MDSSD upsamples multiple high-level feature maps and fuses them with low-level features; it reports strong results across benchmarks, including 77.6% mAP on TT100K.
Problem
Top-most feature maps retain little information about small objects, and upsampling can have difficulty restoring spatial detail lost during feature reduction.
Source: p. 1Method
MDSSD simultaneously upsamples high-level feature maps at multiple scales and fuses them with low-level maps through Fusion Blocks to form Fusion Modules.
Source: pp. 1–2Results
Across small-object and general-detection benchmarks, MDSSD reports strong results, including 77.6% mAP on TT100K and results on VOC2007 ranging from comparable to better depending on the model setup.
Source: p. 3Takeaways & Limitations
Results support MDSSD's effectiveness for small-object detection, with higher average recall for small objects also reported.
Source: p. 3
Abstract
from arXiv · showhide
For most of the object detectors based on multi-scale feature maps, the shallow layers are rich in fine spatial information and thus mainly responsible for small object detection. The performance of small object detection, however, is still less than satisfactory because of the deficiency of semantic information on shallow feature maps. In this paper, we design a Multi-scale Deconvolutional Single Shot Detector (MDSSD), especially for small object detection. In MDSSD, multiple high-level feature maps at different scales are upsampled simultaneously to increase the spatial resolution. Afterwards, we implement the skip connections with low-level feature maps via Fusion Block. The fusion feature maps, named Fusion Module, are of strong feature representational power of small instances. It is noteworthy that these high-level feature maps utilized in Fusion Block preserve both strong semantic information and some fine details of small instances, rather than the top-most layer where the representation of fine details for small objects are potentially wiped out. The proposed framework achieves 77.6% mAP for small object detection on the challenging dataset TT100K with 512 x 512 input, outperforming other detectors with a large margin. Moreover, it can also achieve state-of-the-art results for general object detection on PASCAL VOC2007 test and MS COCO test-dev2015, especially achieving 2 to 5 points improvement on small object categories.