Deep Vision Consulting

Last Saturday, the creator of C++, Bjarne Stroustrup, gave a talk in Florence about concept-based generic programming, and we couldn’t miss the opportunity to attend! Concepts are a C++20 feature introduced to define template requirements more explicitly, improving code readability, providing much clearer error messages, and even speeding up compilation. It was great to hear about this feature directly from the person who designed it, along with a broader overview of generic programming and its potential. This is the best photo Riccardo managed to take (at least you can see Bjarne’s head too 😅). A big thanks to Marco Arena and the Italian C++ Community for making this possible.

Standard 3D annotation tools assume you have LiDAR or depth sensors. Most of the time you don’t. Is there anything you can do to get 3D annotations out of plain 2D cameras, e.g. fixed roadside cameras? Sponsored by the Università degli Studi di Modena e Reggio Emilia, we recently collaborated with AILAB-Udine of Università degli Studi di Udine to solve this. We designed and built SIBORIA, a custom annotation tool that extracts metrically accurate 3D data from purely 2D images. To make this work, we leaned on classical projective geometry. By using natural constraints already in the environment, like road markings, light poles and roundabouts, we solved the depth-scale ambiguity without needing 3D sensors. This calibration allows the tool to generate metrically grounded 3D oriented bounding boxes, which integrate directly with a deep learning auto-labeling model. The result: metrically accurate 3D annotation with no depth sensor and no controlled setup. We wrote a technical post on the math behind this approach and why standard tools fell short. Read it here: https://lnkd.in/gxs9c2K3 — If your company is working with cameras in environments where standard calibration approaches do not apply, whether due to a non-standard mounting position, an absence of depth sensors, or a domain where no existing 3D annotation tool fits, discuss with us what the right approach looks like for your specific context.

With the goal of further consolidating our position as a strategic provider of custom 3D vision solutions, today we are proud to announce the launch of our partnership with Stereolabs, renowned leader for 3D AI vision. In 15 years of field experience, we have consistently seen how 3D data is a decisive factor in many vision challenges. We have processed this data using traditional approaches as well as AI, often deploying it directly on edge devices. We have long recognized the value of stereo vision technology and have used Stereolabs products for nearly a decade, consistently appreciating the quality and reliability of both their hardware and SDK. Building on this solid foundation, the partnership established today between our two companies will further enhance the value delivered to both Deep Vision Consulting and Stereolabs customers. And there is more... as you know, Stereolabs is part of Ouster, whose LiDAR technology Deep Vision Consulting knows intimately and utilizes with great success —a connection that sets the stage for even broader horizons #AI #3D #ComputerVision #StereoCameras #Lidars

At the Renesas Electronics tech day on April 29th in Stuttgart, we'll be showing a vision pipeline with 3 MIPI-CSI cameras on RZ/V2H with DEEPX DX-M1 on an Engicam Srl ev-kit, accelerating image rectification, edge detection and 2 DNN inferences. Let's meet there!

The onsemi iToF sensor is a powerhouse for high-resolution 3D depth mapping. We’ve seen its potential firsthand! Just a heads-up for those needing RGBD: the 'D' makes all the difference and shouldn't be taken for granted, especially if computed at the edge!

Last week Francesco, Andrea and Michele joined 200+ other attendees in Amsterdam to take part in the ELLIS Winter School on Foundation Models. The talks were great and it was super interesting to see how great researchers tackle complex computer vision challenges. But what was most rewarding from a growth perspective was the chance to brainstorm, interact, understand, lose oneself in technical discussions with many other students at an international level. Most of them were pursuing their PhD and were surprised to see people coming from industry, and even more surprised to find out they were that good :) There is no shorter route to excellence than not being satisfied with your current level, this holds both for individual team members and for the company as a whole. ELLIS - European Laboratory for Learning and Intelligent Systems ELLIS unit Amsterdam #AI #ML #computervision #foundationmodels

Today, single-camera, single-model pipelines are commodity, even on embedded devices. Production-grade, multi-camera/multi-model vision pipelines — hardware-accelerated and deployable on efficient edge devices, are definitely a different game.

We are officially opening selections for Autumn '26 internship positions at Deep Vision Consulting 🚀 Looking back, the selection process we held last year brought us two incredible talents: Teresa Campo and Saverio Napolitano. They are currently diving deep into AI-based 3D pose estimation and Generative AI: we are truly proud to have them in our team! Now, it’s your turn. This is your moment to make an impact. Who we’re looking for: - 🎓Master's students in Computer Engineering, Computer Science, or Physics. - 📍Availability to work onsite at our headquarter in Modena. - 📅 Students targeting thesis discussion / graduation around Spring/Summer 2027. What we offer: - 💡Cutting-Edge topics, where advanced research meets real-world application in AI, Computer Vision, and Edge AI. - 💶 An internship grant of up to €750/month. - 👨🏫 Direct guidance from senior experts in the field. Proactive? We also welcome original internship proposals from motivated candidates! Ready to shape the future of Computer Vision with us? Drop us an email at careers@deepvisionconsulting.com or via contact form https://lnkd.in/dFs3rCAM #ComputerVision #AI #DeepLearning #GenerativeAI #Internship #Tirocinio #Modena #TechJobs

What if your next smart camera could detect objects and measure their real-world distance, all on a compact embedded device? As part of the Renesas Preferred Partner Program, we've released an optimized pipeline that turns the Renesas RZ/V2H into what we think edge smart cameras should look like: AI and 3D combined together, all of it computed on a single chip at a price point that makes volume deployment realistic. This is possible because the RZ/V2H goes well beyond the typical SoC with NPU: it adds the DRP, a dedicated image processing accelerator, alongside up to four MIPI-CSI camera inputs, enabling AI and 3D vision to run in parallel, with room for two independent stereo cameras on a single V2H. The pipeline is available on the Renesas RZ/V2H EVK or the Engicam AI-DevKit, which ships with a pre-calibrated stereo camera and our pipeline pre-installed. Come and see it in action at embedded world Exhibition&Conference 2026 at the Renesas and Engicam booths. Thank you to Renesas Electronics for the open and insightful partnership, and to Engicam Srl for making hands-on evaluation of our vision pipeline accessible out of the box. Full technical breakdown on our blog, see link in comments. #EmbeddedWorld2026 #EdgeAI #ComputerVision #SmartCamera #3DVision #StereoCamera

Monte Cimone delivered a genuinely sunny day on the slopes, and going through it with colleagues made it that much better. ❄️