Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces three style-based neural network architectures—Multi-PatchGAN, Truncated ResNet50, and Truncated ResNet50 with Gram Matrix and Attention—for real-time weather classification from images. The Truncated ResNet50 variants, particularly the one incorporating Gram matrices and attention, outperform existing methods by focusing on high-frequency stylistic features. The architectures demonstrate strong generalization across multiple public datasets, suggesting broader applicability to appearance-based classification tasks.
Stripping down ResNet50 and adding Gram matrix attention unlocks surprisingly effective and generalizable image classification, even beating the state-of-the-art in weather detection.
In this paper, we present three neural network architectures designed for real-time classification of weather conditions (sunny, rain, snow, fog) from images. These models, inspired by recent advances in style transfer, aim to capture the stylistic elements present in images. One model, called "Multi-PatchGAN", is based on PatchGANs used in well-known architectures such as Pix2Pix and CycleGAN, but here adapted with multiple patch sizes for detection tasks. The second model, "Truncated ResNet50", is a simplified version of ResNet50 retaining only its first nine layers. This truncation, determined by an evolutionary algorithm, facilitates the extraction of high-frequency features essential for capturing subtle stylistic details. Finally, we propose "Truncated ResNet50 with Gram Matrix and Attention", which computes Gram matrices for each layer during training and automatically weights them via an attention mechanism, thus optimizing the extraction of the most relevant stylistic expressions for classification. These last two models outperform the state of the art and demonstrate remarkable generalization capability on several public databases. Although developed for weather detection, these architectures are also suitable for other appearance-based classification tasks, such as animal species recognition, texture classification, disease detection in medical imaging, or industrial defect identification.