A variation of the code for the paper in our MICCAI'10 paper on stereo reconstruction for MIS is available here. Please note that this is not the original code (it is a simplified version that does the basic propagation) and it does not have the CUDA implementation. It should however run reasonably quickly for VGA resolution images. The code relies on OpenCV (any version from 2.0 upwards should be fine).
Data with ground truth reported in the paper is available from the Hamlyn Centre, Imperial College London website.
Update: Please note that to get the code to work properly the you need to set the value Param.Tt = 200 (or somethings similar). Thanks to Sylvain Bernhardt, UBC for helping to spot this. He has also provided a main.cpp file to demonstrate use of the stereo class.
The scene flow code which is a follow up on the above work and accepted for publication in MICCAI'12 is due soon. Appologies for the delay, I am pretty busy at the moment and packaging it up is taking some time. Will be available in a few weeks.
D. Stoyanov: Stereoscopic Scene Flow for Robotic Assisted Minimally Invasive Surgery. Medical Image Computing and Computer Assisted Interventions (MICCAI12), 479-486, 2012 | | publisher link
Sometime it is useful to create synthetic images of surgial scenes to check the sanity of our algorithms. Obviously this is a limited validation strategy but to overcome the problem of limited ground truth data it is nevertheless useful at times. A long while back I coded a software to do create synthetic data for me. I have been meaning to make it available for some time now but not getting around to it. The executable is somewhat outdated now but will maybe be useful to someone. If you do find it of use, please cite the paper below, thanks!
D. Stoyanov: Stereoscopic Scene Flow for Robotic Assisted Minimally Invasive Surgery. Medical Image Computing and Computer Assisted Interventions (MICCAI12), 2012
We are in the process of releasing data and code for instrument localisation in laparoscopic images. The work is led by Maximilian Allan. At the moment only data data used for training and testing of our classifiers is available but an executable will be available shortly. After a summer clean up, hopefully we can also release the code for download. For a zip of the data have a look at Max's web page.