Lightweight temporal residual modeling for event-based image deblurring

In this paper, we propose a novel framework for event-based image deblurring that integrates perceptual loss and a channel-spatial attention mechanism to enhance visual quality in reconstructed frames. Departing from conventional segment-based temporal modeling, our method employs perceptual supervision via a pretrained ResNet network to prioritize high-level feature consistency over pixel-level accuracy. A Convolutional Block Attention Module (CBAM) further refines feature representation by adaptively weighting informative regions across channel and spatial dimensions, effectively suppressing background noise while preserving critical texture details. The framework is validated on standard event-based benchmarks, with ablation studies confirming the complementary benefits of perceptual supervision and attention mechanisms. By balancing computational efficiency with reconstruction fidelity, the proposed approach demonstrates practical potential for real-time image restoration. Future extensions will focus on adaptive attention strategies and advanced temporal modeling to address more complex motion scenarios.