Open sinsin1998 opened 1 day ago
Updated, thanks for inform me about this
Updated, thanks for inform me about this
Thanks for your quick reply. And I have another question, you said that in your experiment, "the model is trained on eight A10 GPUs, each with 24GB of memory for two days." But I find that 24GB is not enough in your config/OccFusion. Moreover, I don't know that how to use fix_param_small.
Sorry for the confusion. The final version is trained on a few A40s, each with 48 GB, and you can decrease the batch size to fit A10 or 3090/4090. For the fix_param_small just download the fix_param_small from the link and put it inside the OCCFusion folder, the program will automatically load it. You could also set enable_fix=True in the config file to use fix_param_small.
Thanks a lot