TIM: a time interval machine for audio-visual action recognition
<p>Diverse actions give rise to rich audio-visual signals in long videos. Recent works showcase that the two modalities of audio and video exhibit different temporal extents of events and distinct labels. We address the interplay between the two modalities in long videos by explicitly modellin...
| Main Authors: | , , , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
IEEE
2024
|
| _version_ | 1811139184027500544 |
|---|---|
| author | Chalk, J Huh, J Kazakos, E Zisserman, A Damen, D |
| author_facet | Chalk, J Huh, J Kazakos, E Zisserman, A Damen, D |
| author_sort | Chalk, J |
| collection | OXFORD |
| description | <p>Diverse actions give rise to rich audio-visual signals in
long videos. Recent works showcase that the two modalities of audio and video exhibit different temporal extents of
events and distinct labels. We address the interplay between
the two modalities in long videos by explicitly modelling the
temporal extents of audio and visual events. We propose
the Time Interval Machine (TIM) where a modality-specific
time interval poses as a query to a transformer encoder that
ingests a long video input. The encoder then attends to the
specified interval, as well as the surrounding context in both
modalities, in order to recognise the ongoing action.</p>
<p>We test TIM on three long audio-visual video datasets:
EPIC-KITCHENS, Perception Test, and AVE, reporting state-of-the-art (SOTA) for recognition. On EPICKITCHENS, we beat previous SOTA that utilises LLMs and
significantly larger pre-training by 2.9% top-1 action recognition accuracy. Additionally, we show that TIM can be
adapted for action detection, using dense multi-scale interval queries, outperforming SOTA on EPIC-KITCHENS-100
for most metrics, and showing strong performance on the
Perception Test. Our ablations show the critical role of integrating the two modalities and modelling their time intervals in achieving this performance. Code and models at:
https://github.com/JacobChalk/TIM.</p> |
| first_indexed | 2024-09-25T04:02:03Z |
| format | Conference item |
| id | oxford-uuid:9b1cd459-fa33-46ca-b2a6-7911d7e9b408 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-09-25T04:02:03Z |
| publishDate | 2024 |
| publisher | IEEE |
| record_format | dspace |
| spelling | oxford-uuid:9b1cd459-fa33-46ca-b2a6-7911d7e9b4082024-04-24T12:23:44ZTIM: a time interval machine for audio-visual action recognitionConference itemhttp://purl.org/coar/resource_type/c_5794uuid:9b1cd459-fa33-46ca-b2a6-7911d7e9b408EnglishSymplectic ElementsIEEE2024Chalk, JHuh, JKazakos, EZisserman, ADamen, D<p>Diverse actions give rise to rich audio-visual signals in long videos. Recent works showcase that the two modalities of audio and video exhibit different temporal extents of events and distinct labels. We address the interplay between the two modalities in long videos by explicitly modelling the temporal extents of audio and visual events. We propose the Time Interval Machine (TIM) where a modality-specific time interval poses as a query to a transformer encoder that ingests a long video input. The encoder then attends to the specified interval, as well as the surrounding context in both modalities, in order to recognise the ongoing action.</p> <p>We test TIM on three long audio-visual video datasets: EPIC-KITCHENS, Perception Test, and AVE, reporting state-of-the-art (SOTA) for recognition. On EPICKITCHENS, we beat previous SOTA that utilises LLMs and significantly larger pre-training by 2.9% top-1 action recognition accuracy. Additionally, we show that TIM can be adapted for action detection, using dense multi-scale interval queries, outperforming SOTA on EPIC-KITCHENS-100 for most metrics, and showing strong performance on the Perception Test. Our ablations show the critical role of integrating the two modalities and modelling their time intervals in achieving this performance. Code and models at: https://github.com/JacobChalk/TIM.</p> |
| spellingShingle | Chalk, J Huh, J Kazakos, E Zisserman, A Damen, D TIM: a time interval machine for audio-visual action recognition |
| title | TIM: a time interval machine for audio-visual action recognition |
| title_full | TIM: a time interval machine for audio-visual action recognition |
| title_fullStr | TIM: a time interval machine for audio-visual action recognition |
| title_full_unstemmed | TIM: a time interval machine for audio-visual action recognition |
| title_short | TIM: a time interval machine for audio-visual action recognition |
| title_sort | tim a time interval machine for audio visual action recognition |
| work_keys_str_mv | AT chalkj timatimeintervalmachineforaudiovisualactionrecognition AT huhj timatimeintervalmachineforaudiovisualactionrecognition AT kazakose timatimeintervalmachineforaudiovisualactionrecognition AT zissermana timatimeintervalmachineforaudiovisualactionrecognition AT damend timatimeintervalmachineforaudiovisualactionrecognition |