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

The computer vision community has seen a shift from convolutional-based topure transformer architectures for both image and video tasks. Training atransformer from zero for these tasks usually requires a lot of data andcomputational resources. Video Swin Transformer (VST) is a pure-transformermodel developed for video classification which achieves state-of-the-artresults in accuracy and efficiency on several datasets. In this paper, we aimto understand if VST generalizes well enough to be used in an out-of-domainsetting. We study the performance of VST on two large-scale datasets, namelyFCVID and Something-Something using a transfer learning approach fromKinetics-400, which requires around 4x less memory than training from scratch.We then break down the results to understand where VST fails the most and inwhich scenarios the transfer-learning approach is viable. Our experiments showan 85\% top-1 accuracy on FCVID without retraining the whole model which isequal to the state-of-the-art for the dataset and a 21\% accuracy onSomething-Something. The experiments also suggest that the performance of theVST decreases on average when the video duration increases which seems to be aconsequence of a design choice of the model. From the results, we conclude thatVST generalizes well enough to classify out-of-domain videos without retrainingwhen the target classes are from the same type as the classes used to train themodel. We observed this effect when we performed transfer-learning fromKinetics-400 to FCVID, where most datasets target mostly objects. On the otherhand, if the classes are not from the same type, then the accuracy after thetransfer-learning approach is expected to be poor. We observed this effect whenwe performed transfer-learning from Kinetics-400, where the classes representmostly objects, to Something-Something, where the classes represent mostlyactions.

Transfer-learning for video classification: Video Swin Transformer on multiple domains