Deep Learning approaches for multimodal analysis in car collision forensics

Machine Learning and Deep Learning have gained significant traction, playing a pivotal role in addressing a myriad of challenges across various domains. In particular, we decide to exploit them in order to solve numerous tasks as car pose classification, regression of EES (Equivalent Energy Speed), car damage segmentation and 3D re- construction from images and videos. This thesis will present the methods used to solve these problems, along with possible future improvements.