This is a fork of the Anything 3D project for our Deeprob final project. The project allows 3d reconstruction and novel view synthesis from a single monocular image.
In this repository, we included extra setup instructions and slight modifications for easy running. We also implemented a code extension where we test the performance of the project using alternative segmentation models in addition to using GPT-4 for creating text prompts instead of BLIP.
Our project code resides in this directory along with installation instructions. To run this code, you need a NVIDIA CUDA capable machine. We suggest having at least 16GB of vram on GPU.