An implementation using NVIDIA Kaolin Wisp and Stable Diffusion that applies the Text-to-3D model DreamFusion to the task of Few-shot / single-view 3D object reconstruction.
Input RGB views (only 3 sparse views, omitting the right side of the car):
1 | 2 | 3 |
---|---|---|
L2 reconstruction loss on the input views + Score Distillation Sampling with the guiding prompt "a yellow sports car":
car_reconstruction_1.webm
Score Distillation Sampling with a guiding prompt:
1 | 2 | 3 |
---|---|---|
Follow the instrutions here to install NVIDIA Kaolin Wisp.
Run setup.sh
to install the prerequisites.
python main.py --config configs/diffusion_nerf.yaml --dataset-path /path/to/car/dataset --prompt "a car"
python main.py --config configs/diffusion_nerf.yaml --prompt "a DSLR photo of a blue car"