stereo_motorcycle()
The two images are rectified such that every pixel in the left image has its corresponding pixel on the same scanline in the right image. That means that both images are warped such that they have the same orientation but a horizontal spatial offset (baseline). The ground-truth pixel offset in column direction is specified by the included disparity map.
The two images are part of the Middlebury 2014 stereo benchmark. The dataset was created by Nera Nesic, Porter Westling, Xi Wang, York Kitajima, Greg Krathwohl, and Daniel Scharstein at Middlebury College. A detailed description of the acquisition process can be found in .
The images included here are down-sampled versions of the default exposure images in the benchmark. The images are down-sampled by a factor of 4 using the function skimage.transform.downscale_local_mean
. The calibration data in the following and the included ground-truth disparity map are valid for the down-sampled images:
Focal length: 994.978px Principal point x: 311.193px Principal point y: 254.877px Principal point dx: 31.086px Baseline: 193.001mm
The original resolution images, images with different exposure and lighting, and ground-truth depth maps can be found at the Middlebury website .
Left stereo image.
Right stereo image.
Ground-truth disparity map, where each value describes the offset in column direction between corresponding pixels in the left and the right stereo images. E.g. the corresponding pixel of img_left[10, 10 + disp[10, 10]]
is img_right[10, 10]
. NaNs denote pixels in the left image that do not have ground-truth.
Rectified stereo image pair with ground-truth disparities.
Hover to see nodes names; edges to Self not shown, Caped at 50 nodes.
Using a canvas is more power efficient and can get hundred of nodes ; but does not allow hyperlinks; , arrows or text (beyond on hover)
SVG is more flexible but power hungry; and does not scale well to 50 + nodes.
All aboves nodes referred to, (or are referred from) current nodes; Edges from Self to other have been omitted (or all nodes would be connected to the central node "self" which is not useful). Nodes are colored by the library they belong to, and scaled with the number of references pointing them