Dear Qcom Team,
I studied that SNPE support many methods for Image Preprocessing:
- Image Color space conversion
- Image Cropping
- Image Scaling
- Image Mean Subtraction with Image
- Image Mean Subtraction with Channel Value
For the last methond (Image Mean Subtraction with Channel Value), can we use float number for mean_value?
Besides those methods, do we support some thing like Image Mean Multiplication with Channel Value, which is similar to Image Mean Subtraction with Channel Value but use Muliplication Operation ?
Thanks and regards,
CDQ
Q1 : For the last method (Image Mean Subtraction with Channel Value), can we use float number for mean_value?
Ans : As long as the pixel values are normalized, mean_value can take float numbers. In case, if normalization is not preferred, then we cannot use float number for mean_value.
Q2 : Besides those methods, do we support some thing like Image Mean Multiplication with Channel Value, which is similar to Image Mean Subtraction with Channel Value but use Muliplication Operation ?
Ans : Image mean subtraction is a technique for data preparation in order to center the pixel values on the value of zero. This approach is called centering . Using the method of multiplication we cannot achieve this and hence it is not supported in such cases.