analiza aiqbench
This research investigates domain adaptation techniques for motion deblurring models, specifically addressing the performance gap between professional camera data and smartphone imagery. The study finds that input-level conditioning via a severity map yields marginal gains in peak signal-to-noise ratio but does not prove robustness to retraining or establish a definitive spatial-locality mechanism.
Severity-Channel Conditioning: An Injection-Level Study for Consumer-Device Deblurring
arXiv:2610.24599v1 Announce Type: cross Abstract: Motion-deblurring models lose substantial accuracy when transferred from professional cameras to smartphone-derived imagery. This paper studies whether a severity descriptor, partly derived from the paired sharp target, helps adaptation on the iPhoneBlur consumer-device benchmark. Under a shared update budget, input concatenation of a severity map is compared with global feature modulation by capture metadata, spatial feature modulation by the severity map, and a tier-contrastive output objective, together with an unconditioned spectral-loss control, all fine-tuned from the same domain-adapted latent-diffusion baseline. The primary experiment is an oracle diagnostic: target-derived information is supplied at inference. For the evaluated model pair, input concatenation improves overall peak signal-to-noise ratio from 31.23 to 31.30 dB, whereas feature modulation, the contrastive objective, and the spectral-loss control do not exceed the baseline. The gain is positive on 87.2% of test images and on 14 of 15 held-out source videos; this video-level analysis describes the evaluated checkpoint pair and does not establish robustness to retraining, and three training seeds are reported separately. The input-level gain recurs on a transformer backbone. In one exploratory run, zeroing the channel at inference retains most of the gain in peak signal-to-noise ratio but lowers structural similarity below the baseline. A normalized-step diagnostic does not support the expected larger-gradient explanation. The results characterize the evaluated configurations and do not establish injection location as the sole cause, a spatial-locality mechanism, or a reliable deployment advantage.
Abstract:Motion-deblurring models lose substantial accuracy when transferred from professional cameras to smartphone-derived imagery. This paper studies whether a severity descriptor, partly derived from the paired sharp target, helps adaptation on the iPhoneBlur consumer-device benchmark. Under a shared update budget, input concatenation of a severity map is compared with global feature modulation by capture metadata, spatial feature modulation by the severity map, and a tier-contrastive output objective, together with an unconditioned spectral-loss control, all fine-tuned from the same domain-adapted latent-diffusion baseline. The primary experiment is an oracle diagnostic: target-derived information is supplied at inference. For the evaluated model pair, input concatenation improves overall peak signal-to-noise ratio from 31.23 to 31.30 dB, whereas feature modulation, the contrastive objective, and the spectral-loss control do not exceed the baseline. The gain is positive on 87.2% of test images and on 14 of 15 held-out source videos; this video-level analysis describes the evaluated checkpoint pair and does not establish robustness to retraining, and three training seeds are reported separately. The input-level gain recurs on a transformer backbone. In one exploratory run, zeroing the channel at inference retains most of the gain in peak signal-to-noise ratio but lowers structural similarity below the baseline. A normalized-step diagnostic does not support the expected larger-gradient explanation. The results characterize the evaluated configurations and do not establish injection location as the sole cause, a spatial-locality mechanism, or a reliable deployment advantage.
17 pages (12 main + 5 supplementary), 7 figures, 14 tables. Code: this https URL
Image and Video Processing (eess.IV); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Multimedia (cs.MM)
arXiv:2610.24599 [eess.IV]
(or arXiv:2610.24599v1 [eess.IV] for this version)
Text integral sub licența deschisă a publicației · 273 cuvinte