Open INF800 opened 3 years ago
Sorry for later reply, I haven't tried this dataset. According to my experience, self-supervised monocular depth estimation is more suitable for autonomous driving scene, in which the camera only have 3 DOF ( X, Y, YAW).
Hi,
I tried to perform zero-shot depth estimation using intel-midas on panoramic 360deg images (stitched from multiple 2d images). Surprisingly, it does NOT work for panoramic images but works for individual images.
Can I use your method to train on my 360deg panoromic data in self-supervsed way (without any labels) using only the raw images?
Thanks for this amazing project!