Mark Finocchio
Also known as M. Finocchio
Mark Finocchio is a computer-vision researcher whose coauthored pose-recognition work helped bring real-time body tracking to Kinect. In September 2008, he asked Microsoft Research for help with a planned game accessory that could track bodies; in June 2011, he coauthored a CVPR method to predict 3D body joints from one depth image, which won CVPR’s Best Paper Award. He also coauthored a 2011 paper on parallelizing Kinect’s body-parts classifier. His patents cover an on-demand property system and machine-learning classifier training, and CVPR’s 2021 Longuet-Higgins Prize recognizes the pose paper.
Timeline
-
2002–2006
An on-demand property system
In 2002, he files a Microsoft patent for an on-demand property system that tracks dependencies among object properties; the patent is granted in 2006.
-
September 2008
Starting Kinect body tracking
In September 2008, he emails Microsoft Research’s Jamie Shotton to ask for help with a planned game accessory that can track bodies in real time. He is a principal software engineer with Microsoft’s Interactive Entertainment Business.
-
2011
Recognizing pose from depth
At CVPR in June 2011, he coauthors a method that predicts 3D body-joint positions from one depth image by classifying pixels into body parts. The system runs at 200 frames per second on consumer hardware.
-
2011
A CVPR Best Paper Award
The human-pose paper wins CVPR’s Best Paper Award during the 2011 conference.
-
2011
Parallelizing Kinect training
In December 2011, he coauthors a paper on parallelizing the decision-forest training used for Kinect’s body-parts classification system.
-
2016
Partitioning training data
A 2016 patent names him as a co-inventor of a method that uses a global classifier to partition training data into class-specific subsets, then trains a second classifier on those subsets.
-
2021
The Longuet-Higgins Prize
CVPR’s 2021 Longuet-Higgins Prize recognizes the 2011 human-pose paper he coauthors.
Sources and credits
Photo: We have no published photo of Mark Finocchio yet. Send us a free-licensed photo.
- US6986123B2 - Extensible on-demand property system
- Kinect Body Tracking Reaps Renown
- Real-Time Human Pose Recognition in Parts from a Single Depth Image
- Parallelizing the Training of the Kinect Body Parts Labeling Algorithm
- US20160132786A1 - Partitioning data for training machine-learning classifiers
- PAMITC AWARDS | CVPR 2021


