David "Navi" Romero’s Post

Splatting, night powder style...

🔍 In the world of 3D Gaussian Splat output, Luma AI and Postshot are two big players. Our recent comparison of the two revealed some interesting findings. 🚁 We tested footage captured by an FPV drone performing a swift spiral around a church. The results were impressive. Luma AI's output appeared seamless, with no discernible misalignment issues. However, when assessing Postshot's output, we noticed a minor misalignment. 💡 But we have a solution! By pre-aligning videos or photos with photogrammetry software like Reality Capture before importing them into Postshot, we anticipate achieving flawless results. 👀 Stay tuned for our enlightening demonstration in our next post!

Mapping the entire Hong Kong, dated June 23, 2023, covering 2,755 km². Done for Hong Kong Police in under 28 days, covering LOD 1, 2, 3. AI-Accelerated 3D mapping via Satellite Imagery. - Recency vs OUTDATED 3D maps - AI vs manual 3D modelling - Machines vs HUMANs - NEW vs OLD WAYS of modelling cities - $thousands vs $millions of hardware - David vs Goliath #gis #gismapping #live3D #geospatialtechnology #asiapacific #digitaltwins #smartcities #vizzioai #artificialintelligence #generativeai #locationintelligence

The days of LiDAR, run-of-the-mill photogrammetry, and polygon meshes for reality capture and rendering could be numbered.. You may have heard of NeRFs (Neural Radiance Fields), which rely heavily on neural networks to render 3D scenes by examining points from a limited number of photos and generating a "radiance field", but 3D Gaussian Splats, which utilize a more "classical" approach (What is Gaussian Splatting? https://lnkd.in/gjRrHgm5), have recently been taking HUGE leaps forward in quality and fidelity, not to mention that they work exceptionally well for night scenes and render at lightning speeds. And today, there are already apps from Luma AI and PolyCam that you can use to haphazardly capture video for high-quality NeRF and Gaussian Splat scenes using your phone. Lifestyle 360 always has our eye on new tech and are excited to watch these technologies evolve. It's only a matter of time until they mature and become applicable to use in our industry. #realitycapture #radiancefield #gaussiansplatting #rendering

🇬🇧 No more preprocessing delays: CertiflAI will start image control within hours from flight! Get ready, we are gearing up to introduce a new functionality that will speed up aerial image quality control! What's Changing? Soon, we will be rolling out a solution that allows for the seamless processing of IIQ images from PhaseOne cameras, eliminating the need for preprocessing steps for TIF format conversion. Clouds and shadows will be easily detected in the image overview, while blur will be analyzed in the raw green channel—saving you precious time and effort. Why Does It Matter? Time is of the essence in aerial imaging projects. Our latest feature empowers you to swiftly detect and correct crucial image defects, significantly accelerating your project timelines and enhancing quality control. Excited to learn more? Follow us to keep an eye out for updates! #CertiflAI #AerialImaging #AI #Innovation #QualityAssurance 

This is an interesting example of how AI can build 3D maps from simple 2D maps. They may be significantly inaccurate at first, but when we get tons of maps, I believe these algorithms will be able to represent any urban area worldwide with very scarce 2D information.

Topotek 200gram weighted mini size gimbal, DIP290G609, with a fixed focal eo camera and a 9.1 mm 640*512 thermal, also can reach a fabulous AI tracking function. 🎇 As the following video, the gimbal figures out a moving drone from hundreds meters away and keeps tracking on it by gerenal tracking mode. In our AI system, you can also choose "human tracking" or "vehicle tracking" for instead. You can see the trakcing routes clearly from both eo camera and thermal camera. 👏 More details refer to [email protected]

NCCOS is supporting a collaborative mapping project in Lake Superior. #AUVs and drop cameras are used to record video of benthic habitats to help ground-truth our mapping efforts. Knowing the composition of the lakebed (mud, sand, pebbles, cobbles, stones, boulders, etc.,) helps in developing maps and larger predictive models, which in turn is useful for managing the lake's natural resources. Rather than solely relying on visual estimations, our scientists are applying AI to these ground-truthing videos to perform automated object detection, size calculation, and classification. https://lnkd.in/du8y4zrk