high-res-stereo
high-res-stereo copied to clipboard
gengshan-y
Depth Estimation
Hi,
I use your pre-trained model and convert the disp to depth with the following formula,
Depth = f * baseline / pred_disp, where the f and baseline are directly obtained from the calibration file.
When I unproject the depth image to the point cloud, I found the result is no satisfactory.
Do I go sth wrong when calculating the depth map?
Which tool did you use for vis? I thought the cv toolkit uses disparity images as input? It will automatically convert disparity to depth for you.
Which tool did you use for vis? I thought the cv toolkit uses disparity images as input? It will automatically convert disparity to depth for you.
hi，How did you get the camera parameters and correct the image
Camera parameters are provided by the Middlebury dataset: https://vision.middlebury.edu/stereo/
Camera parameters are provided by the Middlebury dataset: https://vision.middlebury.edu/stereo/
The file in high-res-stereo/data-mbtest/0593_8_0661_1/ isn't meant to be a Middlebury dataset, is it? How did Calib.txt come into being?
Pair 0593_8_0661_1 is taken from the "Gangnam Style" sequence in open4d and Spatiotemporal BA. A spatiotemporal bundle adjustment method is used there. Please refer to the paper for details.
I believe classic SfM (e.g. COLMAP) can also be applied to get intrinsics/extrinsics for many use cases.
Hi, Gengshan. Just found you refered the Spatiotemporal BA method. Did you have any reference implementation.
