Example "Train multiclass segmentation on CamVid" Image normalization may be error #1224
Description
in examples/camvid_segmentation_multiclass.ipynb CamVidModel (and other examples) , image is not scaled to [0,1]
def forward(self, image):
# normalize image here
print("param: ", self.mean, self.std)
print("before:", image[0,:,100,100])
image = (image - self.mean) / self.std
print("after:", image[0,:,100,100])
mask = self.model(image)
return mask
and the result is :
param: tensor([[[[0.4850]],
[[0.4560]],
[[0.4060]]]]) tensor([[[[0.2290]],
[[0.2240]],
[[0.2250]]]])
before: tensor([59, 55, 55], dtype=torch.uint8)
after: tensor([255.5240, 243.5000, 242.6400])