AI Paper Atlas

Research map / Video / Video segmentation and tracking

Learning Spatiotemporal Frequency-Transformer for Compressed Video Super-Resolution

Zhongwei Qiu, Huan Yang, Jianlong Fu, Dongmei Fu

ECCV 2022 · Video › Video segmentation and tracking · approach: Real-world video super-resolution · 52 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
transform frames to DCT frequency domain (Pixel+CNN to Frequency+CNN) (ablation)BasicVSR · REDS4 compressed CRF25PSNR: 25.93 -> 26.87Improved
replace CNN with transformer in frequency domain (ablation)Frequency+CNN · REDS4 compressed CRF25PSNR: 26.87 -> 27.15Improved
add frequency attention (FTVSR) over base transformer (ablation)Frequency+Transformer · REDS4 compressed CRF25PSNR: 27.15 -> 27.28Improved
use divided time-space-frequency attention Lambda_ST vs base (ablation)Frequency Transformer · REDS compressed CRF35PSNR: 24.03 -> 24.22Improved
use temporal-only frequency attention vs base (ablation)Frequency Transformer · REDS compressed CRF25PSNR: 27.15 -> 27.11Hurt
use space-frequency attention Lambda_S vs base (ablation)Frequency Transformer · REDS compressed CRF25PSNR: 27.15 -> 27.23Improved
vary DCT block size from 4x4 to 16x16 (ablation)FTVSR Lambda_S · REDS compressed CRF25PSNR: 27.20 -> 27.20No clear effect
use FTVSR instead of COMISRcompressed VSR · REDS4 compressed CRF25PSNR: 26.47 -> 28.05Improved
use FTVSR instead of IconVSRVSR · REDS4 uncompressedPSNR: 31.67 -> 31.82Improved
train FTVSR on both compressed and uncompressed videosFTVSR · REDS4 uncompressedPSNR: 31.82 -> 31.74Hurt
train FTVSR on both compressed and uncompressed videosFTVSR · Vid4 uncompressedPSNR: 28.31 -> 28.06Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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