2025· Advances in Neural Information Processing Systems 38· 0 citations· 58 references
TL;DR
This work pioneers the synergistic integration of SNNs into Transformer architectures for LLIE, establishing a compelling pathway toward powerful, energy-efficient low-level vision on resource-constrained platforms.
Abstract
Low-light image enhancement (LLIE) aims to improve the visibility and quality of images captured under poor illumination. However, existing deep enhancement methods often underemphasize computational efficiency, leading to high energy and memory costs. We propose Spike-RetinexFormer , a novel LLIE architecture that synergistically integrates Retinex theory, spiking neural networks (SNNs) and a Transformer-based design. Leveraging sparse spike-driven computation, the model reduces theoretical compute energy and memory traffic relative to ANN counterparts. Across standard benchmarks, the method matches or surpasses strong ANNs (25.50 dB on LOL-v1; 30.37 dB on SDSD-out) with comparable parameters and lower theoretical energy. Our work pioneers the synergistic integration of SNNs into Transformer architectures for LLIE, establishing a compelling pathway toward powerful, energy-efficient low-level vision on resource-constrained platforms.
This work introduces a Temporal Feedback Coding (TFC) scheme that leverages feedback at the encoding stage to diversify spike patterns and designs a Global-Local Dynamic LIF (GLD-LIF) neuron that enhances cross-step dependency modeling by integrating local aggregation and global initialization.
Experiments on HDRTV1K show that Bio-SFT achieves competitive perceptual quality and consistently improves HDR-VDP-3 and $\Delta E_{ITP}$ while reducing artifact propagation in symmetric guidance pipelines.
Tingyu Cheng, Ting Zhang, Chongyi Li et al.· 0 citations
The haze effect induces light attenuation and color shifting in distant objects, thereby diminishing the contrast and saturation of images. In pursuit of restoring image clarity, a plethora of dehazing methods have been successively introduced. Nevertheless, the development of effective and practical methodologies remains a formidable challenge due to the requirement for powerful feature extraction capability alongside consideration of computational efficiency and model scalability. To address these issues, we put forward a boosted spiking neural network with retina mechanism, termed ReBo-SNN, for the reconstruction of hazy images. Specifically, the proposed network draws inspiration from the antagonistic mechanisms of retinal neurons and innovatively devises a retinal module. By extracting highly expressive features through ON and OFF roads, the retinal module can effectively augment visual resolution and contrast. Furthermore, to circumvent the issue of excessive smoothing, the network introduces a novel boosting strategy that realizes progressive feature enhancement and assists in reconstructing high-quality dehazed images. In addition, the network also incorporates dense residual connections with strided convolution to compensate for potential information loss during the translation, which significantly enhances the efficiency of feature transmission and integration. Extensive experiments on both synthetic and real-world hazy images demonstrate that our proposed network achieves superior quantitative and qualitative results. Based on the spiking mechanism and compact architecture, our network also exhibits a substantial energy efficiency advantage, which promotes its deployment in vision hardware to enhance overall performance.
Ya Zhang, Zhitong Zhao, Qian Sun et al.· Neural Networks· 0 citations
Infrared and visible image fusion (IVIF) integrates the complementary information of two modalities into a single image with richer scene content. While existing methods are largely built on artificial neural networks (ANNs), which densely compute over all activations, spiking neural networks (SNNs) communicate through sparse binary spikes and compute only where and when a spike occurs, offering a route to more energy-efficient fusion. However, directly applying SNNs to IVIF creates a fundamental tension: cross-modal fusion relies on fine-grained responses from both modalities, whereas binary spikes can discard complementary cues that remain below the firing threshold. The membrane potential retains these subthreshold responses before firing, letting both modalities jointly shape the output when integrated at this stage. Building on this, we propose CIS-Fuse, a spiking network that performs cross-modal fusion directly at the membrane-potential level. At its core is the current injection spiking (CIS) operator, which injects one modality as a gated auxiliary current into the driving neuron of the other, so the two integrate before spike firing, with a per-channel learnable injection strength that adaptively regulates the modulation magnitude. Building on CIS, we construct a bidirectional cross-modal fusion (BCMF) module and deploy it on a dual-branch architecture with asymmetric stacking depths, where the two branches develop a clear functional specialization. Extensive experiments on four IVIF benchmarks and on downstream detection and segmentation show that CIS-Fuse achieves fusion quality on par with state-of-the-art ANN-based methods while inheriting the energy efficiency of spike-based computation, with roughly an order of magnitude lower inference energy than the similarly-sized ANN-based DCEvo. Code will be released upon publication.
Rui Zhao, Zhuoyuan Li, Wenrui Li et al.· 0 citations
A Spiking Neural Network (SNN) is a kind of brain-inspired and event-driven network, which is becoming a promising energy-efficient alternative to Artificial Neural Networks (ANNs). In recent years, SNN methods have been successfully applied in the fields of electromagnetic signal processing and image signal processing, particularly in application scenarios that require low energy consumption. However, the performance of SNNs by direct training is far from satisfactory. In this paper, we study a novel learning method named SAD-SNN (Spatial-Activation Distillation for Spiking Neural Networks), which utilizes the ANN model to guide the SNN model learning. Unlike prior works that rely on element-wise feature alignment approaches, SAD-SNN aligns spatial-activation maps at different resolutions of the teacher and student networks. Specifically, we introduce a direct alignment approach, which defines a spatial-activation loss and normalizes the representation vectors of ANN and SNN, to alleviate the unexpected precision loss. This enables the knowledge of teacher ANNs to be effectively transferred to train student SNNs. On three image classification datasets, our proposed SAD-SNN outperforms other SNN training methods no matter whether homogeneous or heterogeneous teacher ANNs are used. Furthermore, we apply SAD-SNN to the electromagnetic signal detection task, demonstrating strong generalization ability and superior performance. In conclusion, the experimental results on various tasks and SNN architectures demonstrate that our method is a general and effective solution that significantly improves the learning of student SNNs with only two time steps.
Chongxiao Qu, Qian Zhang, Chenxiao Dou et al.· Italian National Conference...· 0 citations
Low-light image enhancement (LLIE) aims to improve the visibility and perceptual quality of images captured under insufficient illumination. Recent VQ-based LLIE methods introduce data-driven priors by quantizing normal-light latent features into a vector-quantized (VQ) codebook. However, directly quantizing rich entangled representations often causes severe information loss and requires additional fusion with degraded low-light features, increasing complexity and limiting restoration quality. To address this issue, we build on Retinex theory and propose an Adaptive Retinex Model (ARM) that decomposes an input image into structural (reflectance) and illumination components without relying on handcrafted priors or regularization terms. On top of the illumination branch, we introduce a Global Illumination Representation (GIR) module that learns a VQ codebook of global illumination patterns from normal-light images and uses it as an explicit prior for enhancement. To effectively exploit this illumination codebook for low-light inputs, we design an Illumination Prior Query (IPQ) module and a single-objective query loss, improving the stability and adaptivity of retrieval under the low-/normal-light distribution gap. The resulting framework, termed Adaptive Retinex Network (ARetinex-Net), is extensively evaluated on both synthetic and real-world benchmarks, where it consistently outperforms state-of-the-art LLIE methods in terms of quantitative metrics and visual quality.