Hello author,
I encountered an issue while running evaluation with the nuScenes‑mini dataset. I only made one single modification to the code, as shown below:
alpha = alpha.cpu()
beta = beta.cpu()
gamma = gamma.cpu()
sh_rotations = wigner_D(degree, alpha, beta, gamma).type(dtype)
sh_rotations = sh_rotations.to(device)
I applied this fix to resolve the runtime error caused by tensor device mismatch (some tensors resided on CPU while others were on GPU) during the computation of sh_rotations(gaussian_util.py 199-205). Although the code can now run successfully, the generated results are far from satisfactory. I am wondering whether I have made any mistakes in my operations.
Hello author,
I encountered an issue while running evaluation with the nuScenes‑mini dataset. I only made one single modification to the code, as shown below:
I applied this fix to resolve the runtime error caused by tensor device mismatch (some tensors resided on CPU while others were on GPU) during the computation of
sh_rotations(gaussian_util.py 199-205). Although the code can now run successfully, the generated results are far from satisfactory. I am wondering whether I have made any mistakes in my operations.