Lacunarity Pooling Layers

My M.S. thesis introduced lacunarity pooling, a pooling technique that leverages texture information to improve image classification. I implemented three approaches to lacunarity computation (base, differential box counting, and multi-scale), each integrable with CNN backbones such as ConvNeXt, ResNet18, and DenseNet161. The method reached up to 98.07% accuracy on PlantVillage and 95.00% on LeavesTex1200 with minimal additional parameters.