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

Temporal Reasoning is one important functionality for vision intelligence. Incomputer vision research community, temporal reasoning is usually studied inthe form of video classification, for which many state-of-the-art NeuralNetwork structures and dataset benchmarks are proposed in recent years,especially 3D CNNs and Kinetics. However, some recent works found that currentvideo classification benchmarks contain strong biases towards static features,thus cannot accurately reflect the temporal modeling ability. New videoclassification benchmarks aiming to eliminate static biases are proposed, withexperiments on these new benchmarks showing that the current clip-based 3D CNNsare outperformed by RNN structures and recent video transformers.  In this paper, we find that 3D CNNs and their efficient depthwise variants,when video-level sampling strategy is used, are actually able to beat RNNs andrecent vision transformers by significant margins on static-unbiased temporalreasoning benchmarks. Further, we propose Temporal Fully Connected Block (TFCBlock), an efficient and effective component, which approximates fullyconnected layers along temporal dimension to obtain video-level receptivefield, enhancing the spatiotemporal reasoning ability. With TFC blocks insertedinto Video-level 3D CNNs (V3D), our proposed TFCNets establish newstate-of-the-art results on synthetic temporal reasoning benchmark, CATER, andreal world static-unbiased dataset, Diving48, surpassing all previous methods.

TFCNet: Temporal Fully Connected Networks for Static Unbiased Temporal Reasoning