AI Paper Atlas

Research map / Video / Action recognition

Multimodal Clustering Networks for Self-Supervised Learning From Unlabeled Videos

Brian Chen, Andrew Rouditchenko, Kevin Duarte, Hilde Kuehne, Samuel Thomas, Angie Boggust, Rameswar Panda, Brian Kingsbury, Rogerio Feris, David Harwath, James Glass, Michael Picheny and 1 more

ICCV 2021 · Video › Action recognition · approach: Contrastive video representation learning · 84 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add the clustering loss to MMS contrastive loss (ablation)MCN · YouCook2 R@10R@10: 43.7 -> 44.3Improved
add the clustering loss to MMS contrastive loss (ablation)MCN · MSR-VTT R@10R@10: 32.9 -> 33.7Improved
add the clustering loss to MMS contrastive loss (ablation)MCN · MiningYouTubeIoU: 22.1 -> 22.6Improved
add the reconstruction loss (ablation)MCN · YouCook2 R@10R@10: 44.3 -> 45.2Improved
add the reconstruction loss (ablation)MCN · MiningYouTubeIoU: 22.6 -> 23.1Improved
use MMS instead of NCE as contrastive loss (ablation)MCN · YouCook2 R@10 and MSR-VTT R@10R@10: 39.2 -> 43.7 (YR10); 33.5 -> 32.9 (MR10)Mixed
use MIL-NCE instead of NCE as contrastive loss (ablation)MCN · YouCook2 R@10R@10: 39.2 -> 40.0No clear effect
use centroid target with K-means instead of joint prediction target (ablation)MCN · MiningYouTubeIoU: 21.4 -> 23.1Improved
use joint prediction instead of swap prediction with K-means (ablation)MCN · YouCook2 R@10R@10: 41.3 -> 44.3Improved
use soft instead of hard labels with Sinkhorn swap (ablation)MCN · YouCook2 R@10R@10: 39.0 -> 41.8Improved
add audio modality (retrieve T to VA instead of T to V) (ablation)MCN · YouCook2 R@1R@1: 8.6 -> 18.1Improved
use caption averaging instead of majority vote over clips (ablation)MCN · YouCook2 text-to-full video retrievalR@1: 38.8 -> 53.4Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy