Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs

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

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK USES EXTRA
TRAINING DATA
RESULT BENCHMARK
Semantic Segmentation CamVid DeepLab-MSc-CRF-LargeFOV Mean IoU 61.6% # 10
Real-Time Semantic Segmentation CamVid DeepLab mIoU 61.6% # 15
Time (ms) 203 # 13
Frame (fps) 4.9 # 9
Semantic Segmentation Cityscapes test DeepLab Mean IoU (class) 63.1% # 83
Real-Time Semantic Segmentation Cityscapes test DeepLab mIoU 63.1% # 26
Time (ms) 4000 # 21
Frame (fps) 0.25 # 25
Semantic Segmentation PASCAL VOC 2012 test DeepLab-MSc-CRF-LargeFOV (VGG-16) Mean IoU 71.6% # 41
Scene Segmentation SUN-RGBD DeepLab-LargeFOV Mean IoU 32.08 # 2

Methods used in the Paper


METHOD TYPE
CRF
Structured Prediction
Feedforward Network
Feedforward Networks
Weight Decay
Regularization
SGD with Momentum
Stochastic Optimization
DeepLab
Semantic Segmentation Models
Dropout
Regularization
ReLU
Activation Functions
Max Pooling
Pooling Operations
Convolution
Convolutions
Dense Connections
Feedforward Networks
Softmax
Output Functions
VGG
Convolutional Neural Networks