HTTP/2 301 server: GitHub.com content-type: text/html location: https://mengyuest.github.io/AR-Net/ x-github-request-id: FDBA:3655F2:973042:A9A2DF:69531299 accept-ranges: bytes age: 0 date: Mon, 29 Dec 2025 23:45:30 GMT via: 1.1 varnish x-served-by: cache-bom-vanm7210058-BOM x-cache: MISS x-cache-hits: 0 x-timer: S1767051930.893718,VS0,VE207 vary: Accept-Encoding x-fastly-request-id: abda96e7e703a2422fda6d1693e30aad5412d0ce content-length: 162 HTTP/2 200 server: GitHub.com content-type: text/html; charset=utf-8 last-modified: Mon, 17 Aug 2020 01:49:28 GMT access-control-allow-origin: * etag: W/"5f39e228-2986" expires: Mon, 29 Dec 2025 23:55:30 GMT cache-control: max-age=600 content-encoding: gzip x-proxy-cache: MISS x-github-request-id: 725C:3946E9:95A7BE:A819F8:69531299 accept-ranges: bytes age: 0 date: Mon, 29 Dec 2025 23:45:30 GMT via: 1.1 varnish x-served-by: cache-bom-vanm7210058-BOM x-cache: MISS x-cache-hits: 0 x-timer: S1767051930.114469,VS0,VE215 vary: Accept-Encoding x-fastly-request-id: 17bdc6f0568b3b155b053661477416e0aae11df1 content-length: 3088
AR-Net: Adaptive Frame Resolution for
Efficient Action Recognition

Yue Meng	Chung-Ching Lin	Rameswar Panda	Prasanna Sattigeri

Leonid Karlinsky	Aude Oliva‡	Kate Saenko†	Rogerio Feris

MIT-IBM Watson AI Lab, IBM Research

Boston University†

Massachusetts Institute of Technology‡

ECCV 2020

Abstract

Action recognition is an open and challenging problem in computer vision. While current state-of-the-art models offer excellent recognition results, their computational expense limits their impact for many real-world applications. In this paper, we propose a novel approach, called AR-Net (Adaptive Resolution Network), that selects on-the-fly the optimal resolution for each frame conditioned on the input for efficient action recognition in long untrimmed videos. Specifically, given a video frame, a policy network is used to decide what input resolution should be used for processing by the action recognition model, with the goal of improving both accuracy and efficiency. We efficiently train the policy network jointly with the recognition model using standard back-propagation. Extensive experiments on several challenging action recognition benchmark datasets well demonstrate the efficacy of our proposed approach over state-of-the-art methods.

---

Network

---

Action recognition results on ActivityNet-v1.3 and FCVID

---

Accuracy versus efficiency comparison

---

Paper and Code

		Yue Meng, Chung-Ching Lin, Rameswar Panda, Prasanna Sattigeri, Leonid Karlinsky, Aude Oliva, Kate Saenko, and Rogerio Feris. ARNet: Adaptive Frame Resolution for Efficient Action Recognition European Conference on Computer Vision (ECCV), 2020 [PDF][Code]

---