[QUESTION] Difference between CV-CUDA resize and OpenCV resize

[mzchtx]

We found that the way of calculating coordinate mapping in CV-CUDA's resize is different from that of OpenCV (as shown in the pseudo-code in the figure below):

• OpenCV uses float64, and CV-CUDA uses float32
• OpenCV uses inv_scale = target_size / size, and CV-CUDA uses scale = size / target_size

Due to floating-point calculation errors, the coordinates calculated by OpenCV and CV-CUDA will have a deviation, leading to differences in the final results. For example:

• Python 3: 125 / np.float32(500 / 480) = 120.00000457763689 -> math.floor(120.00000457763689) = 120
• Python 3: 125 / np.float64(500 / 480) = 119.99999999999999 -> math.floor(119.99999999999999) = 119
• CUDA: float32(500 / 480) * 120 = 124.999992 -> __float2int_rd(124.999992) = 124

Does CV-CUDA do this for efficiency?