English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Video semantic segmentation aims to generate accurate semantic maps for eachvideo frame. To this end, many works dedicate to integrate diverse informationfrom consecutive frames to enhance the features for prediction, where a featurealignment procedure via estimated optical flow is usually required. However,the optical flow would inevitably suffer from inaccuracy, and then introducenoises in feature fusion and further result in unsatisfactory segmentationresults. In this paper, to tackle the misalignment issue, we propose aspatial-temporal fusion (STF) module to model dense pairwise relationshipsamong multi-frame features. Different from previous methods, STF uniformly andadaptively fuses features at different spatial and temporal positions, andavoids error-prone optical flow estimation. Besides, we further exploit featurerefinement within a single frame and propose a novel memory-augmentedrefinement (MAR) module to tackle difficult predictions among semanticboundaries. Specifically, MAR can store the boundary features and prototypesextracted from the training samples, which together form the task-specificmemory, and then use them to refine the features during inference. Essentially,MAR can move the hard features closer to the most likely category and thus makethem more discriminative. We conduct extensive experiments on Cityscapes andCamVid, and the results show that our proposed methods significantly outperformprevious methods and achieves the state-of-the-art performance. Code andpretrained models are available at https://github.com/jfzhuang/ST_Memory.

Video Semantic Segmentation with Inter-Frame Feature Fusion and Inner-Frame Feature Refinement