--- title: Bag of Freebies for XR Hand Tracking: Machine Learning &... description: Collabora details the machine learning pipeline behind their open-source XR hand-tracking system, built on the Monado XR platform. The pipeline uses a top-down... canonical: https://daily.dev/posts/bag-of-freebies-for-xr-hand-tracking-machine-learning-openxr-xtoqhisrs twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Bag of Freebies for XR Hand Tracking: Machine Learning & OpenXR | daily.dev og:description: Collabora details the machine learning pipeline behind their open-source XR hand-tracking system, built on the Monado XR platform. The pipeline uses a top-down... og:url: https://daily.dev/posts/bag-of-freebies-for-xr-hand-tracking-machine-learning-openxr-xtoqhisrs og:image: https://api.daily.dev/og/posts/xtOQHISRS.png og:image:alt: Bag of Freebies for XR Hand Tracking: Machine Learning & OpenXR og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Bag of Freebies for XR Hand Tracking: Machine Learning & OpenXR **[Collabora Blog](https://daily.dev/sources/collabora)** · 7 min read · 0 upvotes · 0 comments ## Summary Collabora details the machine learning pipeline behind their open-source XR hand-tracking system, built on the Monado XR platform. The pipeline uses a top-down pose estimation approach: a YOLOv4-inspired CNN detects hands, then an encoder-decoder network predicts 26 keypoints defined in the OpenXR specification using heatmap regression. To address limited labeled training data, the team applied data augmentation strategies and Noisy Student Training — a semi-supervised technique using unlabeled web images — achieving a 7% accuracy improvement and better real-world robustness. The system handles occlusion by incorporating keypoints from previous frames, though extended occlusion and hand-to-object interactions remain open challenges. Source code and pre-trained models are publicly available. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How does Noisy Student Training improve hand pose estimation model accuracy? Noisy Student Training improved model pipeline accuracy by 7% and significantly improved robustness on real-world scenarios. The technique trains a teacher model on labeled data, generates pseudo-labels for unlabeled data, then trains a student model (equal or larger than the teacher) on both labeled and pseudo-labeled data with added noise. The process iterates, with the student becoming the new teacher each round. _ML engineers working on pose estimation pipelines track techniques like these on daily.dev._ ### How many keypoints does the OpenXR specification define for hand tracking? The OpenXR specification defines 26 keypoints for hand tracking. A keypoint estimation network predicts these 26 keypoints using a crop of the image based on a predicted bounding box from a prior hand detection step, outputting a 2D heatmap for each of the 21 predicted keypoints processed via Gaussian regression to calculate final landmarks. _Developers building OpenXR hand-tracking features keep up with spec changes and implementations on daily.dev._ ## Similar posts on daily.dev - [Hands Free, AIs Forward: NVIDIA XR AI Brings Agents to AR Glasses](https://daily.dev/posts/hands-free-ais-forward-nvidia-xr-ai-brings-agents-to-ar-glasses-nin54zara) · NVIDIA · 1 upvotes · 0 comments - [Building AI Agents for AR Glasses and XR Devices with NVIDIA XR AI](https://daily.dev/posts/building-ai-agents-for-ar-glasses-and-xr-devices-with-nvidia-xr-ai-mnrmdlphv) · NVIDIA Developer · 1 upvotes · 0 comments - [Collecting robot training data is dirty, unglamorous work. Some AI labs are already paying XDOF to do it](https://daily.dev/posts/collecting-robot-training-data-is-dirty-unglamorous-work-some-ai-labs-are-already-paying-xdof-to-d-hitmwst6r) · TechCrunch · 0 upvotes · 0 comments - [Explore 3D Objects in Projection Illusion Using Face Tracking](https://daily.dev/posts/explore-3d-objects-in-projection-illusion-using-face-tracking-fzyrq1nkp) · 80 LEVEL · 0 upvotes · 0 comments - [Hands-Free Mouse Uses Eyes And Muscles Instead](https://daily.dev/posts/hands-free-mouse-uses-eyes-and-muscles-instead-1s99qx0am) · Hackaday · 0 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#data-science](https://daily.dev/tags/data-science), [#computer-vision](https://daily.dev/tags/computer-vision) [View this post on daily.dev](https://daily.dev/posts/bag-of-freebies-for-xr-hand-tracking-machine-learning-openxr-xtoqhisrs) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Bag of Freebies for XR Hand Tracking: Machine Learning & OpenXR","url":"https://daily.dev/posts/bag-of-freebies-for-xr-hand-tracking-machine-learning-openxr-xtoqhisrs","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/bag-of-freebies-for-xr-hand-tracking-machine-learning-openxr-xtoqhisrs"},"datePublished":"2026-08-11T08:53:32.761Z","dateModified":"2026-08-11T09:15:13.956Z","description":"Collabora details the machine learning pipeline behind their open-source XR hand-tracking system, built on the Monado XR platform. The pipeline uses a top-down...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/374120b6133dbf9462099b92e60fd20c?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/374120b6133dbf9462099b92e60fd20c?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Collabora Blog","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collabora Blog","logo":"https://media.daily.dev/image/upload/s--ZGZ0FjqS--/f_auto,q_auto/v1786438306/logos/collabora?_a=BAMAMicg0","url":"https://daily.dev/sources/collabora"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/bag-of-freebies-for-xr-hand-tracking-machine-learning-openxr-xtoqhisrs","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,data-science,computer-vision","timeRequired":"PT7M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collabora Blog","item":"https://daily.dev/sources/collabora"},{"@type":"ListItem","position":3,"name":"Bag of Freebies for XR Hand Tracking: Machine Learning & OpenXR"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/bag-of-freebies-for-xr-hand-tracking-machine-learning-openxr-xtoqhisrs#faq","mainEntity":[{"@type":"Question","name":"How does Noisy Student Training improve hand pose estimation model accuracy?","acceptedAnswer":{"@type":"Answer","text":"Noisy Student Training improved model pipeline accuracy by 7% and significantly improved robustness on real-world scenarios. The technique trains a teacher model on labeled data, generates pseudo-labels for unlabeled data, then trains a student model (equal or larger than the teacher) on both labeled and pseudo-labeled data with added noise. The process iterates, with the student becoming the new teacher each round. ML engineers working on pose estimation pipelines track techniques like these on daily.dev."}},{"@type":"Question","name":"How many keypoints does the OpenXR specification define for hand tracking?","acceptedAnswer":{"@type":"Answer","text":"The OpenXR specification defines 26 keypoints for hand tracking. A keypoint estimation network predicts these 26 keypoints using a crop of the image based on a predicted bounding box from a prior hand detection step, outputting a 2D heatmap for each of the 21 predicted keypoints processed via Gaussian regression to calculate final landmarks. Developers building OpenXR hand-tracking features keep up with spec changes and implementations on daily.dev."}}]} ```