The dataset I'm working on at CVLab will be based in the famous "Head and Lamp" stereo pair, created at University of Tsukuba in 1997.
For this dataset, we are planning to create a big CG generated scene, and a one-minute-long video, rendered inside Maya with a stereo camera.
Thanks to the CG technologies, we can obtain a 100% accuracy in the depth and disparity maps, that are impossible to obtain with a real life camera nowadays.
This will provide us with 1800 stereo pairs dataset with 100% accurate ground truth data, that we believe will be very useful for future researchers in the field of computer vision.
In addition, our dataset could also be used for other lines of investigation, such as camera tracking algorithms, among others...
Here you can see some sample images of the original stereo dataset, the model in wire view and the rendered model with Fluorescent Lights illumination:
No comments :
Post a Comment