Fix MegaDepth train/val split

[akretz]

The __init__() function of the MegaDepth data loader has a split parameter. However, it never gets passed to the super initializer and so self.split never gets set. This means self.split is always None for the MegaDepth dataset and training data = validation data.

dust3r/dust3r/datasets/megadepth.py

Lines 16 to 29 in 4c24a6e

	class MegaDepth(BaseStereoViewDataset):
	def __init__(self, *args, split, ROOT, **kwargs):
	self.ROOT = ROOT
	super().__init__(*args, **kwargs)
	self.loaded_data = self._load_data(self.split)
	if self.split is None:
	pass
	elif self.split == 'train':
	self.select_scene(('0015', '0022'), opposite=True)
	elif self.split == 'val':
	self.select_scene(('0015', '0022'))
	else:
	raise ValueError(f'bad {self.split=}')