Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs

22 Dec 2014Liang-Chieh ChenGeorge PapandreouIasonas KokkinosKevin MurphyAlan L. Yuille

Deep Convolutional Neural Networks (DCNNs) have recently shown state of the art performance in high level vision tasks, such as image classification and object detection. This work brings together methods from DCNNs and probabilistic graphical models for addressing the task of pixel-level classification (also called "semantic image segmentation")... (read more)

PDF Abstract

Evaluation results from the paper


Task Dataset Model Metric name Metric value Global rank Compare
Semantic Segmentation CamVid DeepLab-MSc-CRF-LargeFOV Mean IoU 61.6% # 6
Real-Time Semantic Segmentation CamVid DeepLab mIoU 61.6% # 5
Real-Time Semantic Segmentation CamVid DeepLab Time (ms) 203 # 3
Real-Time Semantic Segmentation CamVid DeepLab Frame (fps) 4.9 # 3
Semantic Segmentation Cityscapes DeepLab Mean IoU (class) 63.1% # 23
Real-Time Semantic Segmentation Cityscapes DeepLab mIoU 63.1% # 9
Real-Time Semantic Segmentation Cityscapes DeepLab Time (ms) 4000 # 10
Real-Time Semantic Segmentation Cityscapes DeepLab Frame (fps) 0.25 # 12
Semantic Segmentation PASCAL VOC 2012 DeepLab-MSc-CRF-LargeFOV Mean IoU 71.6% # 15
Scene Segmentation SUN-RGBD DeepLab-LargeFOV Mean IoU 32.08 # 1