Corresponding points between 2D image(s) and 3D object(s). Stored in JSON files typically called obj_img_cors.json.
An n-object-m-image correspondence point is the coordinates of one same point on n 3D objects, and m 2D images.
For example the position of a scene point in world space (X, Y, Z), and the positions of the pixels showning that point on two images (ix1, iy1) and (ix2, iy2), would be a 1-object 2-images correspondence.
n-object-m-image correspondences are an array of n-object-m-image correspondence points. For all the points, the relative arrangement of the object(s) and image camera(s) are the same.
For example capturing a chessboard pattern from two cameras generates 1-object-2-image correspondences with points for all the checkerboard corners; their pixel positions on both images, the their 3D position on the checkerboard’s own coordinate system.
Finally, a n-object-m-image correspondences set is an array of the latter. (That is, an array of arrays of correspondence points).
The relative arrangement of object(s) and image camera(s) can be different on the outer array. And they can be different images.
For example when a checkerboard is captured multiple times with two cameras, the output data is a 1-object-2-image correspondences set. The placement of the object (the checkerboard) relative to the cameras is different for each image.
kinect/checkerboard_samples generates a 1-object-2-image correspondences set with chessboard corners, from images taken with a Kinect’s color and IR camera.
misc/cat_obj_img_cors is used to merge multiple object-image correspondences sets together.