Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

End-to-End Learning of Visual Representations from Uncurated Instructional Videos

About

Annotating videos is cumbersome, expensive and not scalable. Yet, many strong video models still rely on manually annotated data. With the recent introduction of the HowTo100M dataset, narrated videos now offer the possibility of learning video representations without manual supervision. In this work we propose a new learning approach, MIL-NCE, capable of addressing misalignments inherent to narrated videos. With this approach we are able to learn strong video representations from scratch, without the need for any manual annotation. We evaluate our representations on a wide range of four downstream tasks over eight datasets: action recognition (HMDB-51, UCF-101, Kinetics-700), text-to-video retrieval (YouCook2, MSR-VTT), action localization (YouTube-8M Segments, CrossTask) and action segmentation (COIN). Our method outperforms all published self-supervised approaches for these tasks as well as several fully supervised baselines.

Antoine Miech, Jean-Baptiste Alayrac, Lucas Smaira, Ivan Laptev, Josef Sivic, Andrew Zisserman• 2019

Related benchmarks

TaskDatasetResultRank
Action RecognitionUCF101
Accuracy91.3
365
Action RecognitionUCF101 (mean of 3 splits)
Accuracy91.3
357
Text-to-Video RetrievalMSR-VTT
Recall@19.9
313
Action RecognitionUCF101 (test)--
307
Text-to-Video RetrievalMSR-VTT (test)
R@1990
234
Action RecognitionHMDB51
Top-1 Acc61
225
Text-to-Video RetrievalMSR-VTT (1k-A)
R@1032.4
211
Action RecognitionHMDB-51 (average of three splits)
Top-1 Acc61
204
Text-to-Video RetrievalMSRVTT (test)
Recall@19.9
155
Action RecognitionUCF101 (3 splits)
Accuracy91.3
155
Showing 10 of 91 rows
...

Other info

Code

Follow for update