Waleed Abdulla

Engineering Leader, GenAI and 3D perception

I lead an applied research organization at Apple, building and growing multiple teams that develop GenAI and Computer Vision algorithms behind key features of Vision Pro and other Apple products.

My interests include generative models, LLMs, object tracking, hand tracking, computer graphics, and synthetic data.

Previously, I worked on Computer Vision and Machine Learning projects across Silicon Valley startups. I created two open source projects that reached tens of thousands of users.

Earlier, I founded and led a venture-backed startup that I scaled to over 7 million users, and served on the board of Hacker Dojo, a non-profit technology community center in Mountain View.

My most significant and recent work is at Apple (2018–present) and is confidential. Below are links to related public Apple materials, followed by selected open-source projects from earlier in my career.

I can't detail my specific contributions, but my organization develops the kinds of algorithms behind the capabilities described in these materials. They're shared as context for my work, not as claims about any individual feature.

visionOS 26 introduces powerful new spatial experiences for Apple Vision Pro

Apple Newsroom: overview of the new spatial experiences and capabilities introduced in visionOS 26.

What's new in visionOS 26

WWDC25 session: technical overview of the visionOS 26 release for developers.

Explore spatial accessory input on visionOS

WWDC25 session: developer guide to spatial accessory input on visionOS.

Explore object tracking for visionOS

WWDC24 session: developer guide to object tracking on visionOS.

Tracking and visualizing hand movement

Apple Developer documentation: guide to tracking and visualizing hand movement on visionOS.

Mask R-CNN for Object Detection and Segmentation

Star Fork

An implementation of Mask RCNN on Keras and TensorFlow. I built this during my work at Matterport and they graciously agreed to open source it. Since its release in November 2017, it has become one of the top instance segmentation models on TensorFlow and was used by thousands of developers in applications ranging from Kaggle competitions to Ph.D theses. It has earned 984 academic citations, plus 25K+ stars and 11K+ forks on Github.

https://github.com/matterport/Mask_RCNN

HiddenLayer: Neural Network Visualization and Monitoring

Star Fork

Like TensorBoard, but simpler and runs directly in Jupyter notebook. It was adopted by Microsoft as one of the building blocks of their TensorWatch tool. Supports PyTorch, TensorFlow, and Keras. It has 1,864 stars and 268 forks on Github.

https://github.com/waleedka/hiddenlayer

Splash of Color: Instance Segmentation with Mask R-CNN and TensorFlow (2018)

A question I kept getting about my Mask RCNN implementation was: how do I use it with my own dataset? So I built a toy example and documented the whole process, including manually annotating a small dataset from scratch. The result was an article and code to automatically apply a color splash effect on images of balloons (i.e. convert the image to grayscale except for balloons, which stay colored).

Link to article

Traffic Sign Recognition with TensorFlow (2017)

I occasionally talk about deep learning at local events. This is a talk I did for ACM (Association for Computing Machinery) about using TensorFlow to recognize traffic signs.

Recording on YouTube