Guided Upsampling Network for Real-Time Semantic Segmentation

Mazzini, Davide

doi:10.48550/arxiv.1807.07466

Cited by 21 publications

(26 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To enhance the ability of spatial invariance in CNN, Jaderberg et al [28] introduce a new learnable module to explicitly allow the spatial manipulation of data or feature. GUM [29] proposed a guided upsampling module to learn 2D transformation offsets for each position. Our approach shares a similar aspect with the three approaches [28], [29], [30] in that we also learn 2D transformation offsets.…”

Section: Related Workmentioning

confidence: 99%

“…GUM [29] proposed a guided upsampling module to learn 2D transformation offsets for each position. Our approach shares a similar aspect with the three approaches [28], [29], [30] in that we also learn 2D transformation offsets. But unlike these three approaches, our goal is to align multiresolution features from different convolutional blocks for better aggregation, and thus our approach requires to take features from different blocks as inputs rather than from a single block in [28].…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

AlignSeg: Feature-Aligned Segmentation Networks

Huang

Wei

Wang

et al. 2020

Preprint

View full text Add to dashboard Cite

Aggregating features in terms of different convolutional blocks or contextual embeddings has been proven to be an effective way to strengthen feature representations for semantic segmentation. However, most of current popular network architectures tend to ignore the misalignment issues during the feature aggregation process caused by 1) step-by-step downsampling operations, and 2) indiscriminate contextual information fusion. In this paper, we explore the principles in addressing such feature misalignment issues and inventively propose Feature-Aligned Segmentation Networks (AlignSeg). AlignSeg consists of two primary modules, i.e., the Aligned Feature Aggregation (AlignFA) module and the Aligned Context Modeling (AlignCM) module. First, AlignFA adopts a simple learnable interpolation strategy to learn transformation offsets of pixels, which can effectively relieve the feature misalignment issue caused by multiresolution feature aggregation. Second, with the contextual embeddings in hand, AlignCM enables each pixel to choose private custom contextual information in an adaptive manner, making the contextual embeddings aligned better to provide appropriate guidance. We validate the effectiveness of our AlignSeg network with extensive experiments on Cityscapes and ADE20K, achieving new state-of-the-art mIoU scores of 82.6% and 45.95%, respectively. Our source code will be made available.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

AlignSeg: Feature-Aligned Segmentation Networks

Huang

Wei

Wang

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…ShelfNet18-lw (CVPR2019) [57] -74.8 95 36.9 @ 1080 Ti 36.9 -SwiftNet-RN18 (CVPR2019) [39] 75.4 -104.0 39.9 @ 1080 Ti 39.9 -DABNet (BMVC2019) [33] 70.1 --27.7 @ 1080 Ti 27.7 -BiSeNet-RN18 (ECCV2018) [32] 74.8 74.7 67 65.5 @ Titan Xp 58.5 -ICNet (ECCV2018) [10] -69.5 30 30.3 @ Titan X 49.7 -GUNet (BMVC2018) [58] -70.4 -33.3 @ Titan Xp 29.7 -ERFNet (TITS2017) [11] 72.7 69.7 27.7 11.2 @ Titan X 18.4 -…”

Section: Efficient Static Networkmentioning

confidence: 99%

SegBlocks: Towards Block-Based Adaptive Resolution Networks for Fast Segmentation

Verelst

Tuytelaars

2020

Lecture Notes in Computer Science

View full text Add to dashboard Cite

SegBlocks reduces the computational cost of existing neural networks, by dynamically adjusting the processing resolution of image regions based on their complexity. Our method splits an image into blocks and downsamples blocks of low complexity, reducing the number of operations and memory consumption. A lightweight policy network, selecting the complex regions, is trained using reinforcement learning. In addition, we introduce several modules implemented in CUDA to process images in blocks. Most important, our novel BlockPad module prevents the feature discontinuities at block borders of which existing methods suffer, while keeping memory consumption under control. Our experiments on Cityscapes and Mapillary Vistas semantic segmentation show that dynamically processing images offers a better accuracy versus complexity trade-off compared to static baselines of similar complexity. For instance, our method reduces the number of floating-point operations of SwiftNet-RN18 by 60% and increases the inference speed by 50%, with only 0.3% decrease in mIoU accuracy on Cityscapes.

show abstract

“…Designing an organized network structure is one of the keys to achieving high segmentation accuracy. Popular real-time segmentation structures include encoder-decoder networks [9], [10], [11], [12], [13], multi-scale networks [4], [10], [14], [15], [16], multi-branch networks [17], [18], [19], and multi-stage networks [20], [21], etc. Examples of these structures are visualized in Figure 1.…”

Section: Introductionmentioning

confidence: 99%

CSRNet: Cascaded Selective Resolution Network for Real-time Semantic Segmentation

Xiong¹,

Po²,

Yu³

et al. 2021

Preprint

View full text Add to dashboard Cite

Real-time semantic segmentation has received considerable attention due to growing demands in many practical applications, such as autonomous vehicles, robotics, etc. Existing real-time segmentation approaches often utilize feature fusion to improve segmentation accuracy. However, they fail to fully consider the feature information at different resolutions and the receptive fields of the networks are relatively limited, thereby compromising the performance. To tackle this problem, we propose a light Cascaded Selective Resolution Network (CSRNet) to improve the performance of real-time segmentation through multiple context information embedding and enhanced feature aggregation. The proposed network builds a three-stage segmentation system, which integrates feature information from low resolution to high resolution and achieves feature refinement progressively. CSR-Net contains two critical modules: the Shorted Pyramid Fusion Module (SPFM) and the Selective Resolution Module (SRM). The SPFM is a computationally efficient module to incorporate the global context information and significantly enlarge the receptive field at each stage. The SRM is designed to fuse multiresolution feature maps with various receptive fields, which assigns soft channel attentions across the feature maps and helps to remedy the problem caused by multi-scale objects. Comprehensive experiments on two well-known datasets demonstrate that the proposed CSRNet effectively improves the performance for real-time segmentation.

show abstract

Guided Upsampling Network for Real-Time Semantic Segmentation

Cited by 21 publications

References 0 publications

AlignSeg: Feature-Aligned Segmentation Networks

AlignSeg: Feature-Aligned Segmentation Networks

SegBlocks: Towards Block-Based Adaptive Resolution Networks for Fast Segmentation

CSRNet: Cascaded Selective Resolution Network for Real-time Semantic Segmentation

Contact Info

Product

Resources

About