txytju / det

单目标检测

一、相关论文

序号	时间	论文名字	会议	git
1	2015.12.8	SSD Deconvolutional Single Shot Detector	ECCV2016
2	2017.1.23	DSSD Deconvolutional Single Shot Detector
3	2017.5.26	R-SSD Enhancement of SSD by concatenating feature maps for object detection
4	2017.07.18	RUN：Residual Features and Unified Prediction Network for Single Stage Detection		git
5	2017.7.26	Detecting Small Signs from Large Images
6	2017.8.17	S3FD Single Shot Scale-invariant Face Detector
7	2016.8.24	Online Real-time Multiple Spatiotemporal Action Localisation and Prediction	ICCV 2017	git
8	2017.8.27	Context-aware single-shot detector
9	2017.9.15	Feature-Fused SSD: Fast Detection for Small Objects
10	2017.11.18	Single-Shot Refinement Neural Network for Object Detection		git
11	2017.11.27	Receptive Field Block Net for Accurate and Fast Object Detection		git
12	2017.12.1	Single-Shot Object Detection with Enriched Semantics
13	2017.12.4	FSSD: Feature Fusion Single Shot Multibox Detector
14	2017.12.8	Weaving Multi-scale Context for Single Shot Detector

二、相关方法对比

训练条件:VOC07+12
对比条件: (T):Titanx Maxwell (P):Titan X Pascal GPU Speed(P)=2*Speed(T) 以下数据全部来自论文统计

序号	方法	输入大小	基础网络	速度fps	mAP（VOC07)	mAP（VOC12)	备注
1	SSD	300	VGG16	46	77.2	75.8
2	DSSD	300	Resnet-101	9.5	78.6	76.3
3	RefineDet	320	VGG16	40.3	80	78.1
4	DES	300	VGG16	34	79.5	77	67.8(P)
5	DSOD	300	DS/64-192-48-1	17.4	77.7	-
6	R-SSD	300	VGG16	37.1	78.5	-
7	RUN3WAY	300	VGG16	40	79.2
8	FSSD	300	VGG16	33	78.8	-	65.6(P)
9	DiCSSD	300	VGG16	40.8	78.1	-	-
10	DeCSSD	300	VGG16	39.8	77.6
11	Proposed element-sum model	300	VGG16	43	78.9
12	RFBNet	300	VGG16	43	80.5
13	RFBNet	300	VGG16	83	80.5
-	-
序号	方法	输入大小	基础网络	速度fps	mAP（VOC07)	mAP（VOC12)	备注
1	SSD	512	VGG16	19	79.8	78.5
2	DSSD	513	Resnet-101	5.5	81.5	80
3	RefineDet	512	VGG16	24.1	81.8	80.1
4	DES	512	VGG16	14	81.6	80.2	27.2(P)
5	R-SSD	512	VGG16	15.8	80.8
6	RUN3WAY	512	VGG16	19.5	80.9
7	FSSD	512	VGG16	18	80.9	-	35.7(P)
8	RFB	512	VGG16	18	82.2
9	RFBNet	512	VGG16	38	82.23