I have a depth map reconstruction problem that I have formulated in this form :
With w(u,v) is the weighting term.
To produce the two maps R and W I used a convolutional neural network.
Now, I performed tests (grid search) to find the optimal (
) term that minimizes the reconstruction error.
With this I used the same number of test images (200) to test the precision and recall of the CNN to produce images against ground truth data.
Is the amount of test data sufficient to perform this kind of evaluation? If so, why?
I note that the input images contain 3D objects that we want to reconstruct through this process.
