memory usage

[lfz]

VISTA/monailabel/monaivista/lib/model/vista_point_2pt5/inferer.py

Line 94 in 1e7f3cc

pred_volume = torch.repeat_interleave(torch.zeros_like(inputs_l), num_classes + 1, dim=1).float()

this line use too much memory when image is big

the definition of variable "pred_volume“ varies in different functions. it is 104 channel prob logits at beginning, but changed to 1 channel segmentation finally. To reduce memory use, it should be claimed as 1 channel segmentation map