SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation

We present a novel and practical deep fully convolutional neural network architecture for semantic pixel-wise segmentation termed SegNet. This core trainable segmentation engine consists of an encoder network, a corresponding decoder network followed by a pixel-wise classification layer... (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 ADE20K SegNet Validation mIoU 21.64 # 23
Lesion Segmentation Anatomical Tracings of Lesions After Stroke (ATLAS) SegNet Dice 0.2767 # 8
IoU 0.1911 # 7
Precision 0.3938 # 8
Recall 0.2532 # 8
Semantic Segmentation CamVid SegNet Mean IoU 46.4% # 12
Real-Time Semantic Segmentation CamVid SegNet mIoU 46.4% # 16
Time (ms) 217 # 14
Frame (fps) 4.6 # 10
Real-Time Semantic Segmentation Cityscapes test SegNet mIoU 57.0% # 30
Time (ms) 60 # 16
Frame (fps) 16.7 # 20
Semantic Segmentation Cityscapes test SegNet Mean IoU (class) 57.0% # 88
Medical Image Segmentation RITE SegNet Dice 52.23 # 3
Jaccard Index 39.14 # 2
Semantic Segmentation SkyScapes-Dense SegNet Mean IoU 23.14 # 6
Scene Segmentation SUN-RGBD SegNet Mean IoU 31.84 # 3

Results from Other Papers


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK USES EXTRA
TRAINING DATA
SOURCE PAPER COMPARE
Crowd Counting UCF-QNRF Encoder-Decoder MAE 270 # 11

Methods used in the Paper


METHOD TYPE
Convolution
Convolutions
Kaiming Initialization
Initialization
Batch Normalization
Normalization
ReLU
Activation Functions
Max Pooling
Pooling Operations
Softmax
Output Functions
SGD with Momentum
Stochastic Optimization
SegNet
Semantic Segmentation Models