With reference to the original MNIST paper image (Figure 1), the results obtained from the example of STN of matconvnet (Figure 2) results in the focusing on the object without any apparent transformations as shown in Figure 2 . Is that the final result after the affine transformation or do we need to use the matrix of affine to transform the image ourselves. Infact, I have tried using the 6 affine parameters using the matlab's affine2d and imwarp command but to no use.
Hi,
With reference to the original MNIST paper image (Figure 1), the results obtained from the example of STN of matconvnet (Figure 2) results in the focusing on the object without any apparent transformations as shown in Figure 2 . Is that the final result after the affine transformation or do we need to use the matrix of affine to transform the image ourselves. Infact, I have tried using the 6 affine parameters using the matlab's affine2d and imwarp command but to no use.
As a reference, http://openaccess.thecvf.com/content_ECCV_2018/papers/Sergio_Silva_License_Plate_Detection_ECCV_2018_paper.pdf , this paper does the same thing i am looking for which is implemented on tensorflow. Is it possible to achieve the same using matconvnet.
Thanks in advance.