def create_and_check_for_pretraining(self, config, pixel_values, labels):
model = ViTMAEForPreTraining(config)
model.to(torch_device)
model.eval()
result = model(pixel_values)
# expected sequence length = num_patches
image_size = to_2tuple(self.image_size)
patch_size = to_2tuple(self.patch_size)
num_patches = (image_size[1] // patch_size[1]) * (image_size[0] // patch_size[0])
expected_seq_len = num_patches
expected_num_channels = self.patch_size ** 2 * self.num_channels
self.parent.assertEqual(result.logits.shape, (self.batch_size, expected_seq_len, expected_num_channels))
def create_and_check_for_pretraining(self, config, pixel_values, labels):
model = ViTMAEForPreTraining(config)
model.to(torch_device)
model.eval()
result = model(pixel_values)
num_patches = (self.image_size // self.patch_size)**2
expected_num_channels = self.patch_size**2 * self.num_channels
self.parent.assertEqual(
result.logits.shape,
(self.batch_size, num_patches, expected_num_channels))
# test greyscale images
config.num_channels = 1
model = ViTMAEForPreTraining(config)
model.to(torch_device)
model.eval()
pixel_values = floats_tensor(
[self.batch_size, 1, self.image_size, self.image_size])
result = model(pixel_values)
expected_num_channels = self.patch_size**2
self.parent.assertEqual(
result.logits.shape,
(self.batch_size, num_patches, expected_num_channels))