AI Paper Atlas

Research map / Video / Action recognition

Self-Conditioned Probabilistic Learning of Video Rescaling

Yuan Tian, Guo Lu, Xiongkuo Min, Zhaohui Che, Guangtao Zhai, Guodong Guo, Zhiyong Gao

ICCV 2021 · Video › Action recognition · approach: Contrastive video representation learning · 36 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add temporal modeling by switching Dense2D to Dense2D-T (ablation)SelfC-basic · Vid4-YPSNR-Y: 30.62 -> 31.29Improved
increase GMM mixture number from 1 to 5 (ablation)SelfC-basicT · Vid4-YPSNR-Y: 31.29 -> 31.61Improved
enlarge backbone 4 times (2 to 8 Dense2D-T blocks) (ablation)SelfC-small · Vid4-YPSNR-Y: 31.61 -> 32.30Improved
add self-conditioned STP-Net probability model to auto-encoder (ablation)Auto-Enc · Vid4-YPSNR-Y: 28.91 -> 31.61Improved
replace Dense2D backbone with temporal Dense2D-T in IRN (ablation)IRN · Vid4-YPSNR-Y: 30.68 -> 30.42Hurt
use SelfC instead of IRN for rescalingSelfC-large · Vid4PSNR-Y: 31.29 -> 32.30Improved
use proposed control-variates gradient estimator instead of straight-through estimator for H.265 (ablation)SelfC compression · video compressionPSNR: nearly 0.3dBImproved
add SelfC rescaling before H.265 (default mode)H.265 · UVGBDBR-SSIM: -50.84Improved
add SelfC rescaling before H.265 (zerolatency mode)H.265 · HEVC Class BBDBR-SSIM: -18.04Improved
use learned SelfC downscaler instead of bicubic, no fine-tuningTSM · Something V1Top-1 accuracy: 30.4 -> 32.7Improved
use learned downscaler instead of bicubic with action CNN fine-tunedTSM · Something V1Top-1 accuracy: 42.1 -> 43.5Improved
add end-to-end joint fine-tuning of the downscaler (ablation)TSM with SelfC · Something V1Top-1 accuracy: 43.5 -> 44.6Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy