DefaultSegmentorV2

class DefaultSegmentorV2(num_classes, backbone_out_channels, backbone=None, criteria=None, freeze_backbone=False, mlp_head=False)

Bases: Module

train(mode=True)

Set module mode while keeping a frozen probe backbone in eval mode.

nn.Module.train() recurses into every child module. The trainer calls it at the start of each epoch, so setting the backbone to eval mode only in __init__ is not enough: BatchNorm buffers would keep updating and dropout/drop-path would remain active during a nominal linear probe.

DefaultSegmentorV2.forward(input_dict, return_point=False)