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

Deep video recognition is more computationally expensive than imagerecognition, especially on large-scale datasets like Kinetics [1]. Therefore,training scalability is essential to handle a large amount of videos. In thispaper, we study the factors that impact the training scalability of videonetworks. We recognize three bottlenecks, including data loading (data movementfrom disk to GPU), communication (data movement over networking), andcomputation FLOPs. We propose three design guidelines to improve thescalability: (1) fewer FLOPs and hardware-friendly operator to increase thecomputation efficiency; (2) fewer input frames to reduce the data movement andincrease the data loading efficiency; (3) smaller model size to reduce thenetworking traffic and increase the networking efficiency. With theseguidelines, we designed a new operator Temporal Shift Module (TSM) that isefficient and scalable for distributed training. TSM model can achieve 1.8xhigher throughput compared to previous I3D models. We scale up the training ofthe TSM model to 1,536 GPUs, with a mini-batch of 12,288 video clips/98,304images, without losing the accuracy. With such hardware-aware model design, weare able to scale up the training on Summit supercomputer and reduce thetraining time on Kinetics dataset from 49 hours 55 minutes to 14 minutes 13seconds, achieving a top-1 accuracy of 74.0%, which is 1.6x and 2.9x fasterthan previous 3D video models with higher accuracy. The code and more detailscan be found here: http://tsm-hanlab.mit.edu.

Training Kinetics in 15 Minutes: Large-scale Distributed Training on Videos