Cityscapes Dataset - Preparing depth data for deep learning (Part 2 - CreStereo)
Learn how to leverage Cre-Stereo to create high quality depth maps for Deep Learning.
At least 16bits are required to fully represent the depth map, make sure to save in the appropriate format, a .tiff or numpy (.npy) will work, a .png as shown in the video will truncate to an unsigned 8 bit image.
Part 1: • Cityscapes Dataset - Preparing depth data ...
Cityscapes dataset: https://www.cityscapes-dataset.com/
Cityscapes paper: https://arxiv.org/pdf/1604.01685.pdf
Cre-Stereo Pytorch: https://github.com/ibaiGorordo/CRESte...
notebook: https://github.com/itberrios/CREStere...