Wyniki wyszukiwania - BazTech

Ograniczanie wyników

2 Machine Graphics and Vision

Powiadomienia systemowe

Sesja wygasła!

Znaleziono wyników: 2

Liczba wyników na stronie

Wyniki wyszukiwania

Sortuj według:

Ogranicz wyniki do:

Perceptually Optimised Swin-Unet for Low-Light Image Enhancement

Lehmann Tomasz M., Rokita Przemysław

Machine Graphics and Vision

2025

Vol. 34, No. 4

23--42

In this paper we propose a novel approach to low-light image enhancement using a transformer-based Swin-Unet and a perceptually driven loss that incorporates Learned Perceptual Image Patch Similarity (LPIPS), a deep-feature distance aligned with human visual judgements. Specifically, our U-shaped Swin-Unet applies shifted-window self-attention across scales with skip connections and multi-scale fusion, mapping a low-light RGB image to its enhanced version in one pass. Training uses a compact objective - Smooth-L1, LPIPS (AlexNet), MS-SSIM (detached), inverted PSNR, channel-wise colour consistency, and Sobel-gradient terms - with a small LPIPS weight chosen via ablation. Our work addresses the limits of purely pixel-wise losses by integrating perceptual and structural components to produce visually superior results. Experiments on LOL-v1, LOL-v2, and SID show that while our Swin-Unet does not surpass current state-of-the-art on standard metrics, the LPIPS-based loss significantly improves perceptual quality and visual fidelity. These results confirm the viability of transformer-based U-Net architectures for low-light enhancement, particularly in resource-constrained settings, and suggest exploring larger variants and further tuning of loss parameters in future work.

Attention-based U-Net for image demoiréing

Lehmann Tomasz M.

Machine Graphics and Vision

2022

Vol. 31, No. 1/4

3--17

Image demoiréing is a particular example of a picture restoration problem. Moiré is an interference pattern generated by overlaying similar but slightly offset templates. In this paper, we present a deep learning based algorithm to reduce moiré disruptions. The proposed solution contains an explanation of the cross-sampling procedure - the training dataset management method which was optimized according to limited computing resources. Suggested neural network architecture is based on Attention U-Net structure. It is an exceptionally effective model which was not proposed before in image demoiréing systems. The greatest improvement of this model in comparison to U-Net network is the implementation of attention gates. These additional computing operations make the algorithm more focused on target structures. We also examined three MSE and SSIM based loss functions. The SSIM index is used to predict the perceived quality of digital images and videos. A similar approach was applied in various computer vision areas. The author’s main contributions to the image demoiréing problem contain the use of the novel architecture for this task, innovative two-part loss function, and the untypical use of the cross-sampling training procedure.