CUDA out of memory even if max prim is reduced

[biervat]

I still get the error out of memory at around 43%. I tried to reduce args.max_prim. I tried a lot of values even as low as 1 but it doesn't seem to change anything it keeps going out of memory at the same time.

[565353780]

Just edit the function in the Dataset class:

def filter_smallset(self):
    ...
    # "training -> train"
    if self.split == "train":
        ...
    else:
        # add self.max_prim >= len(target) for val and test data
        if self.max_prim >= len(target) >= self.filter_num:
        ...

[avinash-218]

Image size = 100 max_prim = 1200 and the below

def filter_smallset(self):
    ...
    # "training -> train"
    if self.split == "train":
        ...
    else:
        # add self.max_prim >= len(target) for val and test data
        if self.max_prim >= len(target) >= self.filter_num:
        ...

did not rectify the CUDA error

[idoglanz]

I think the issue might be coming from the definition of the CADDataLoader class: def __init__(self, split='train', do_norm=True, cfg=None, max_prim=12000): where max_prim is supplied and used as a separate argument (not as part of the cfg object where it is being updated per the external arg)

Long story short, either send it separately when using the class: train_dataset = CADDataLoader(split='train', do_norm=cfg.do_norm, cfg=cfg, max_prim=cfg.max_prim) or set the self.max_prim = cfg.max_prim in line 17 in dataset.py