I am trying to implement visual odometry using stereo images. In order to do this, stereo parameters of the camera are required. The camera parameters are obtained by calibrating the camera with a checkerboard. I have found many sample datasets of stereo images online but none of them provide the camera parameters of the images. What I don't understand is, what is the point of uploading the stereo images without providing the parameters of the camera that was used to capture the images. Am I missing something?
Given two images of a static scene without a calibration you can only reconstruct the scene up to a projective ambiguity (see Hartley & Zisserman chapter 9.5). To estimate the camera parameters you need at least three views. Maybe the datasets you found contain more than two images of the same scene ? Using multiple (usually a lot more than two) views, it is possible to calculate the calibration and reconstruct the scene.