jzbontar/mc-cnn

what is the preprocessing?

sooyeonshin opened this issue · 3 comments

sooyeonshin commented

can i ask you about your work?

I don't understand about preprocessing with kitti data set.

their are several outs exist like x0.bin, x1.bin, metadata, nnz_tr, nnz_te...

look's like x0.bin is the image pixel data and metadata is the image imformation.
but I can't understand what is tr, te, nnz_tr and nnz_te.

nnz_tr = torch.FloatTensor(23e6, 4)?

what is 23e6 meaning?

jzbontar commented
  • te: a list of size 40. Contains the indices of the validation set.
  • tr: a list of size 154 (194 - 40). Contains the indices of the training set.
  • nnz_{tr, te}: Contains the coordinates at which the ground truth is defined (ground truth disparity is not available at all locations of an image). The array is populated in the make_dataset2 function. 23e6 is just a big number. I know there are not more than 23e6 ground truth disparities (I have an assert for that in the C code).
  • 👍1
Qurey commented

@sooyeonshin @jzbontar
I don't understand about preprocessing with middlebury data set
in the preprocess_mb.py
mask = cv2.imread('tmp/mask.png', 0)
disp0[mask != 255] = 0
y, x = np.nonzero(mask == 255).
what is the function of mask and what is mask meaning?

jzbontar commented

The mask stores information about which areas are occluded. Try opening the generated mask.png and it will make sense.