Due to computational resource limitations, we are not able to process and train our model on the full Objaverse dataset.

To test the out-of-domain generalization ability, we generate images of imaginary objects as the input to our model (see Fig. 10).