DepthLux: Depthwise Separable Convolution Transformer

Low-light image enhancement is a critical task in computer vision, often challenged by noise, low contrast, and color distortion. Traditional convolutional neural networks rely heavily on normal convolutions, which can be computationally intensive and less efficient in handling spatial dependencies in low light conditions. We propose a novel transformer-based framework for low-light image enhancement that eschews nor-mal convolutions in favor of depthwise separable convolutions. Experimental results demonstrate that our method matches the performance of state-of-the-art methods by adopting a novel approach for the feed forward network, which typically uses linear and convolutional filters that increase computational cost. This approach presents a promising direction for efficient and effective low-light image enhancement in real-world applications.