I'm now running CLIP similarity on this implementation and noticed that there's a significant difference in the 3D object results.
Since you mentioned in the readme that there are "slight hyperparameter differences," I'm curious about how I can tune them back to the original GaussianDreamer parameters.
Nice work!
I'm now running CLIP similarity on this implementation and noticed that there's a significant difference in the 3D object results.
Since you mentioned in the readme that there are "slight hyperparameter differences," I'm curious about how I can tune them back to the original GaussianDreamer parameters.
prompt: a 20-sided die made out of glass
Original GaussianDreamer https://github.com/user-attachments/assets/abc00d47-3d82-4076-bb7a-9a9221001512
This implementation https://github.com/user-attachments/assets/570213a8-eace-4087-97f5-4eb29de70763