Querying in images using bilinear interpolation


I have a input T1 of size (1,256,256,3) i.e. images/grid of batch size 1. I have another tensor T2 of size (1, N, 2) i.e. tensor consisting of coordinates i.e. [ [10.5 , 200.787], [150.568, 190.456], …]. How do I compute functional values (using bilinear interpolation) of coordinates in T2 from T1 data?

Or is there any function equivalent to tensorflow function “tf.contrib.resampler.resampler”?

Thanks for any help


1 Like

You would have to normalize your coordinates to [-1, 1], permute the input to have the shape [batch_size, channels, height, width], and you could apply grid_sample.

1 Like

Hi prtblck,

Thanks for the reply. I have used it as you have suggested
The output of grid_sample however is not as expected. I am attaching the files input image, grid and output.
Output is being replicated four times after i apply grid sample as visible in the image input_image grid_ output
Order of images: Input image(left), grid(right), output(bottom).
May I know where would it go wrong?