3D Image Classification Based on Multi View CNN Using 2D Images
Premanand Pralhad Ghadekar, Mrugakshi Deshmukh, Shreyash Deshmukh, Devansh Jangid, Dhanashree Dewalkar, Rushikesh Dighole · 2023
Classifying 2D images using a 2D image dataset is less taxing. Detecting 3D objects efficiently is often a challenge since it requires a decent amount of computational power. In order to remedy this niche shortcoming, this model attempts to achieve the same result, i.e., classifying 3D objects using a 2D image dataset. By doing so the computational complexity can be reduced greatly. This makes the entire pipeline of classification substantially more streamlined. Taking the 2D inputs the 3D object is created using Mesh and Coordinate method and then the model is fed to the Multiview Convolutional Neural Network model for detection purposes. This approach of classification can be used in applications such as robotics to classify 3D objects in the environment from 2D camera inputs. The other applications can be augmented and virtual reality, e-commerce, etc.