%0 Generic %1 journals/corr/abs-1902-04502 %A Poudel, Rudra P K %A Liwicki, Stephan %A Cipolla, Roberto %D 2019 %K arch backbone segmentation ssd %T Fast-SCNN: Fast Semantic Segmentation Network %U http://arxiv.org/abs/1902.04502 %X The encoder-decoder framework is state-of-the-art for offline semantic image segmentation. Since the rise in autonomous systems, real-time computation is increasingly desirable. In this paper, we introduce fast segmentation convolutional neural network (Fast-SCNN), an above real-time semantic segmentation model on high resolution image data (1024x2048px) suited to efficient computation on embedded devices with low memory. Building on existing two-branch methods for fast segmentation, we introduce our `learning to downsample' module which computes low-level features for multiple resolution branches simultaneously. Our network combines spatial detail at high resolution with deep features extracted at lower resolution, yielding an accuracy of 68.0% mean intersection over union at 123.5 frames per second on Cityscapes. We also show that large scale pre-training is unnecessary. We thoroughly validate our metric in experiments with ImageNet pre-training and the coarse labeled data of Cityscapes. Finally, we show even faster computation with competitive results on subsampled inputs, without any network modifications.

@misc{journals/corr/abs-1902-04502, abstract = {The encoder-decoder framework is state-of-the-art for offline semantic image segmentation. Since the rise in autonomous systems, real-time computation is increasingly desirable. In this paper, we introduce fast segmentation convolutional neural network (Fast-SCNN), an above real-time semantic segmentation model on high resolution image data (1024x2048px) suited to efficient computation on embedded devices with low memory. Building on existing two-branch methods for fast segmentation, we introduce our `learning to downsample' module which computes low-level features for multiple resolution branches simultaneously. Our network combines spatial detail at high resolution with deep features extracted at lower resolution, yielding an accuracy of 68.0% mean intersection over union at 123.5 frames per second on Cityscapes. We also show that large scale pre-training is unnecessary. We thoroughly validate our metric in experiments with ImageNet pre-training and the coarse labeled data of Cityscapes. Finally, we show even faster computation with competitive results on subsampled inputs, without any network modifications.}, added-at = {2019-03-12T22:25:23.000+0100}, author = {Poudel, Rudra P K and Liwicki, Stephan and Cipolla, Roberto}, biburl = {https://www.bibsonomy.org/bibtex/2514be181cbc6a235729665edc4d74f7d/nmatsuk}, description = {Fast-SCNN: Fast Semantic Segmentation Network}, interhash = {2816994824c5043d1d0f657a5f79ff6c}, intrahash = {514be181cbc6a235729665edc4d74f7d}, keywords = {arch backbone segmentation ssd}, note = {cite arxiv:1902.04502}, timestamp = {2019-03-12T22:25:23.000+0100}, title = {Fast-SCNN: Fast Semantic Segmentation Network}, url = {http://arxiv.org/abs/1902.04502}, year = 2019 }