Massively parallel video networks
We introduce a class of causal video understanding models that aims to improve efficiency of video processing by maximising throughput, minimising latency, and reducing the number of clock cycles. Leveraging operation pipelining and multi-rate clocks, these models perform a minimal amount of computa...
| Main Authors: | , , , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
Springer, Cham
2018
|
| _version_ | 1797052392348319744 |
|---|---|
| author | Carreira, J Patraucean, V Mazare, L Zisserman, A Osindero, S |
| author_facet | Carreira, J Patraucean, V Mazare, L Zisserman, A Osindero, S |
| author_sort | Carreira, J |
| collection | OXFORD |
| description | We introduce a class of causal video understanding models that aims to improve efficiency of video processing by maximising throughput, minimising latency, and reducing the number of clock cycles. Leveraging operation pipelining and multi-rate clocks, these models perform a minimal amount of computation (e.g. as few as four convolutional layers) for each frame per timestep to produce an output. The models are still very deep, with dozens of such operations being performed but in a pipelined fashion that enables depth-parallel computation. We illustrate the proposed principles by applying them to existing image architectures and analyse their behaviour on two video tasks: action recognition and human keypoint localisation. The results show that a significant degree of parallelism, and implicitly speedup, can be achieved with little loss in performance. |
| first_indexed | 2024-03-06T18:30:58Z |
| format | Conference item |
| id | oxford-uuid:09a63482-0b82-4764-a999-38158a8be875 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-06T18:30:58Z |
| publishDate | 2018 |
| publisher | Springer, Cham |
| record_format | dspace |
| spelling | oxford-uuid:09a63482-0b82-4764-a999-38158a8be8752022-03-26T09:19:28ZMassively parallel video networksConference itemhttp://purl.org/coar/resource_type/c_5794uuid:09a63482-0b82-4764-a999-38158a8be875EnglishSymplectic ElementsSpringer, Cham2018Carreira, JPatraucean, VMazare, LZisserman, AOsindero, SWe introduce a class of causal video understanding models that aims to improve efficiency of video processing by maximising throughput, minimising latency, and reducing the number of clock cycles. Leveraging operation pipelining and multi-rate clocks, these models perform a minimal amount of computation (e.g. as few as four convolutional layers) for each frame per timestep to produce an output. The models are still very deep, with dozens of such operations being performed but in a pipelined fashion that enables depth-parallel computation. We illustrate the proposed principles by applying them to existing image architectures and analyse their behaviour on two video tasks: action recognition and human keypoint localisation. The results show that a significant degree of parallelism, and implicitly speedup, can be achieved with little loss in performance. |
| spellingShingle | Carreira, J Patraucean, V Mazare, L Zisserman, A Osindero, S Massively parallel video networks |
| title | Massively parallel video networks |
| title_full | Massively parallel video networks |
| title_fullStr | Massively parallel video networks |
| title_full_unstemmed | Massively parallel video networks |
| title_short | Massively parallel video networks |
| title_sort | massively parallel video networks |
| work_keys_str_mv | AT carreiraj massivelyparallelvideonetworks AT patrauceanv massivelyparallelvideonetworks AT mazarel massivelyparallelvideonetworks AT zissermana massivelyparallelvideonetworks AT osinderos massivelyparallelvideonetworks |