2020 AVA Challenge: results, talks & reports

[David Ross]

Hi Everyone,

I'm happy to announce that the full results of the 2020 AVA challenge are now available on the website: http://research.google.com/ava/challenge.html.

Congratulations to the team from CUHK-SenseTime, who got first place in action detection on AVA-Kinetics with an impressive 39.62 mAP, and increase of over +5 mAP from last year's strong result.

Congratulations also to the team from Universidad de los Andes for winning the ActiveSpeaker task with an mAP of 86.68!

• Reports from the top teams are linked from the Challenge page.
• The video of the results presentation, and presentations from the two winning teams is available on YouTube as well as on bilibili.
• I also gave a talk in the Large-scale Holistic Video Understanding Tutorial at CVPR covering how context & attention has been used to improve performance on AVA over the past 3 years. Check it out on YouTube.

Thanks to everyone who entered, and to the organizers of the International Challenge on Activity Recognition (ActivityNet) workshop who hosted the challenge.

See you next year!

David Ross

Google Research