CUDA out of memery

[LHXhh]

I test unbounded dataset garden in the train_mlp_nerf, which works well during training process, but there is something wrong during testing. Is that normal? Thank you very much.

[liruilong940607]

Simply because the GPU you are using doesn't not have enough GPU memory to support this task. Testing would require extra memory.

[liruilong940607]

You can use smaller target_sample_batch_size to save memory for training, or chunk_size to save memory for inference. Or sample less points via increasing render_step_size

[LHXhh]

I will have a try, Thank you.