Real Time Semantic Segmentation On Cityscapes

评估指标

Frame (fps)

mIoU

评测结果

各个模型在此基准测试上的表现结果

			Paper Title
PIDNet-L	31.1(3090)	80.6%	PIDNet: A Real-time Semantic Segmentation Network Inspired by PID Controllers
SFNet-R18	25.7(1080Ti)	80.4%	Semantic Flow for Fast and Accurate Scene Parsing
PIDNet-M	42.2(3090)	79.8%	PIDNet: A Real-time Semantic Segmentation Network Inspired by PID Controllers
PIDNet-S	93.2(3090)	78.6%	PIDNet: A Real-time Semantic Segmentation Network Inspired by PID Controllers
RegSeg (no ImageNet pretraining)	30	78.3%	Rethinking Dilated Convolution for Real-time Semantic Segmentation
PP-LiteSeg-B2	102.6(1080Ti)	77.5%	PP-LiteSeg: A Superior Real-Time Semantic Segmentation Model
DDRNet-23-slim	101.6(2080Ti)	77.4%	Deep Dual-resolution Networks for Real-time and Accurate Semantic Segmentation of Road Scenes
STDC2-75	97.0(1080Ti)	76.8%	Rethinking BiSeNet For Real-time Semantic Segmentation
SwinMTL	-	76.41%	SwinMTL: A Shared Architecture for Simultaneous Depth Estimation and Semantic Segmentation from Monocular Camera Images
HyperSeg-M	36.9	75.8%	HyperSeg: Patch-wise Hypernetwork for Real-time Semantic Segmentation
SwiftNetRN-18	39.9	75.5%	In Defense of Pre-trained ImageNet Architectures for Real-time Semantic Segmentation of Road-driving Images
BiSeNet V2-Large	47.3	75.3%	BiSeNet V2: Bilateral Network with Guided Aggregation for Real-time Semantic Segmentation
STDC1-75	126.7	75.3%	Rethinking BiSeNet For Real-time Semantic Segmentation
TD4-BISE18	47.6 (Titan X)	74.9%	Temporally Distributed Networks for Fast Video Semantic Segmentation
PP-LiteSeg-T2	143.6(1080Ti)	74.9%	PP-LiteSeg: A Superior Real-Time Semantic Segmentation Model
ShelfNet18	59.2	74.8%	ShelfNet for Fast Semantic Segmentation
BiSeNet(ResNet-18)	65.5	74.7%	BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation
BiSeNet	65.5	74.7%	BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation
PP-LiteSeg-B1	195.3(1080Ti)	73.9%	PP-LiteSeg: A Superior Real-Time Semantic Segmentation Model
SegBlocks-RN18 (t=0.4)	48.6 (1080Ti)	73.8%	SegBlocks: Block-Based Dynamic Resolution Networks for Real-Time Segmentation

0 of 39 row(s) selected.