Multi-Modal Earth Observation and Deep Learning for Urban Scene Understanding

📌 This repository is a collection of all sub-repositories each of which has been been developed to solve a specific objective set by this research. The repositories are organized as follows: LabelNoise: Implementatoin to simulate of label noise and then prepare the dataset for training and comparison. SegHead: Efficient semantic segmentation Head with lower memory footprint and compute requirement without comprimising model performance. TransFusion: Implementation TranFusion, a transformer based multimodal network for fusing Image and 3D point cloud further improving 2D semaantic segmentation. SOTA-Test: This repository contains implementations of various SOTA models for quantitative comparison and benchmarking purposes.