I need to replicate PyTorch image normalization in OpenCV or NumPy.
Quick backstory: I’m doing a project where I’m training in PyTorch but will have to inference in OpenCV due to deploying to an embedded device where I won’t have the storage space to install PyTorch. Therefore, I need to use NumPy to do the normalization before inferencing on device.
I’m getting different accuracy results when normalizing and test inferencing in PyTorch vs normalizing in NumPy and inferencing in OpenCV, and I suspect the difference is due to the normalization process producing a slightly different result between the two.
Here is a quick script I put together to show the difference on a single image. Note that I’m using grayscale (single-channel) and I’m normalizing into the -1.0 to +1.0 range:
# scratchpad.py
import torch
import torchvision
import cv2
import numpy as np
import PIL
from PIL import Image
TRANSFORM = torchvision.transforms.Compose([
torchvision.transforms.Resize((224, 224)),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.5], [0.5])
])
def main():
# 1st show PyTorch normalization
# open the image as an OpenCV image
openCvImage = cv2.imread('image.jpg', cv2.IMREAD_GRAYSCALE)
# convert OpenCV image to PIL image
pilImage = PIL.Image.fromarray(openCvImage)
# convert PIL image to a PyTorch tensor
ptImage = TRANSFORM(pilImage).unsqueeze(0)
# show the PyTorch tensor info
print('\nptImage.shape = ' + str(ptImage.shape))
print('ptImage max = ' + str(torch.max(ptImage)))
print('ptImage min = ' + str(torch.min(ptImage)))
print('ptImage avg = ' + str(torch.mean(ptImage)))
print('ptImage: ')
print(str(ptImage))
# 2nd show NumPy normalization
# resize the image
openCvImage = cv2.resize(openCvImage, (224, 224))
# convert to float 32 (necessary for passing into cv2.dnn.blobFromImage which is not show here)
openCvImage = openCvImage.astype('float32')
mean = np.mean(openCvImage)
stdDev = np.std(openCvImage)
openCvImage = (openCvImage - mean) / stdDev
# show results
print('\nopenCvImage.shape = ' + str(openCvImage.shape))
print('openCvImage max = ' + str(np.max(openCvImage)))
print('openCvImage min = ' + str(np.min(openCvImage)))
print('openCvImage avg = ' + str(np.mean(openCvImage)))
print('openCvImage: ')
print(str(openCvImage))
print('\ndone !!\n')
# end function
if __name__ == '__main__':
main()
Here is the test image that I’m using:
and here are the results I’m getting currently:
$ python3 scratchpad.py
ptImage.shape = torch.Size([1, 1, 224, 224])
ptImage max = tensor(0.9608)
ptImage min = tensor(-0.9686)
ptImage avg = tensor(0.1096)
ptImage:
tensor([[[[ 0.0431, -0.0431, 0.1294, ..., 0.8510, 0.8588, 0.8588],
[ 0.0510, -0.0510, 0.0980, ..., 0.8353, 0.8510, 0.8431],
[ 0.0588, -0.0431, 0.0745, ..., 0.8510, 0.8588, 0.8588],
...,
[ 0.6157, 0.6471, 0.5608, ..., 0.6941, 0.6627, 0.6392],
[ 0.4902, 0.3961, 0.3882, ..., 0.6627, 0.6471, 0.6706],
[ 0.3725, 0.4039, 0.5451, ..., 0.6549, 0.6863, 0.6549]]]])
openCvImage.shape = (224, 224)
openCvImage max = 2.1724665
openCvImage min = -2.6999729
openCvImage avg = 7.298528e-09
openCvImage:
[[ 0.07062991 -0.42616782 0.22349077 ... 1.809422 1.8476373
1.809422 ]
[ 0.10884511 -0.42616782 -0.04401573 ... 1.7520993 1.7903144
1.7520993 ]
[ 0.0324147 -0.42616782 -0.21598418 ... 1.809422 1.8285296
1.8285296 ]
...
[ 1.1597633 1.5419154 1.0642253 ... 1.7712069 1.178871
1.1406558 ]
[ 1.4081622 0.56742764 0.70118093 ... 1.3890547 1.3699471
1.3126242 ]
[ 0.56742764 0.7393961 1.0069026 ... 1.1024406 1.5419154
1.178871 ]]
So I’m getting somewhat close results to the PyTorch normalization but definitely not the same.
How can I do the normalization in NumPy and have it come out exactly or almost exactly the same as the PyTorch normalization?
Admittedly this is a semi-repost of this question on Stack Overflow python - How to replicate PyTorch normalization in OpenCV or NumPy? - Stack Overflow but I did not get a working answer there so I figured I’d ask here.