.png){kind=icon}
.png){kind=icon}
Return
E2L-CAMixerSR: Early-to-late group summary fusion for lightweight image super-resolution in edge vision systems
.png){kind=icon}
OA
AI
C
Q
Z
J
DOI：10.1016/j.icte.2026.06.006.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Lightweight image super-resolution is important for resource-constrained edge vision systems, where reconstruction quality must be improved under limited memory, computation, and bandwidth budgets. Although CAMixerSR provides an efficient content-aware framework, it lacks explicit low-cost information reuse between the early and late stages of reconstruction. We propose E2L-CAMixerSR, which introduces early-to-late group summary fusion to inject compact pooled summaries from early groups into later groups as channel-wise structural guidance. The proposed design preserves the efficiency advantages of CAMixerSR while adding only 0.73% parameters and negligible FLOP overhead. Across ×2 , ×3 , and ×4 benchmarks, E2L-CAMixerSR improves the average PSNR over the baseline by 0.0127 dB, 0.0243 dB, and 0.0264 dB, respectively, with the clearest gain on Urban100 at ×4 (26.0837 dB versus 25.9989 dB). These results show that selective early-to-late summary reuse is an effective lightweight design for edge-oriented super-resolution.
Keywords:
Edge vision systems
Group summary fusion
Image super-resolution
Lightweight neural networks
Selective feature reuse
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
4.2
Papers:
960
Citations:
2.5K
Organization
No organization information available
Cited Papers