Albert Olivier’s Post

"Metric Flow Matching for Smooth Interpolations on the Data Manifold" by Kacper Kapusniak, Peter Potaptchik, Teodora Reu, Leo Zhang, Alexander Tong, Michael Bronstein, Avishek Bose and Francesco Di Giovanni "Matching objectives underpin the success of modern generative models and rely on constructing conditional paths that transform a source distribution into a target distribution. Despite being a fundamental building block, conditional paths have been designed principally under the assumption of Euclidean geometry, resulting in straight interpolations. However, this can be particularly restrictive for tasks such as trajectory inference, where straight paths might lie outside the data manifold, thus failing to capture the underlying dynamics giving rise to the observed marginals. In this paper, we propose Metric Flow Matching (MFM), a novel simulation-free framework for conditional flow matching where interpolants are approximate geodesics learned by minimizing the kinetic energy of a data-induced Riemannian metric. This way, the generative model matches vector fields on the data manifold, which corresponds to lower uncertainty and more meaningful interpolations. We prescribe general metrics to instantiate MFM, independent of the task, and test it on a suite of challenging problems including LiDAR navigation, unpaired image translation, and modeling cellular dynamics. We observe that MFM outperforms the Euclidean baselines, particularly achieving SOTA on single-cell trajectory prediction." Paper: https://lnkd.in/d92zV5i7 #machinelearning

Leverage #Hexagon's largest database of temporally consistent, very-high-resolution #aerial imagery and #elevation data of the U.S. and Europe 🌎 ! Hexagon #PixelDelivery provides imagery and elevation data for local use over the area of your choice. Perfect for long-term projects or projects requiring local data for analysis, data is seamless and ready for analysis. All imagery is orthorectified with 30 cm resolution coverage of the United States and parts of Europe, and 15 cm resolution in metro areas (cities larger than 50,000 population) of the United States! If you only need a very-high-resolution reference layer for visualization purposes, then Hexagon #Streaming Service is your solution! The perfect #basemap for #GIS and #mapping professionals needing accuracy at an affordable price, which can be inserted directly into your GIS or enterprise application of choice, including Esri, MapInfo, and Global Mapper. ➡️ Find out more about Hexagon Pixel Delivery: https://lnkd.in/dS3Z6BbZ ➡️ Find out more about Hexagon Streaming Service: https://lnkd.in/dYeTMa2p 💬 Can't decide which product suits best to your project? Check out our #comparison tool: https://lnkd.in/epevD2yF

As geospatial professionals, we do everything we can to minimize both spatial and nonspatial error so that we deliver high quality products that deliver value to our customers. However, when it comes to #GeoAI, we do not have a set of best practices and standards like what American Society for Photogrammetry and Remote Sensing (ASPRS) publishes. Have a look at what I am talking about here: https://lnkd.in/eKid-XNp Our industry needs guidance based on sound science. I loathe the word interdisciplinary, but the measurement pros need to talk to the machine learning folks and try to agree on something and get that pushed out to the public. Should we assume our models have the same spatial accuracy as that of the coarsest dataset fed to our models? Does the ordering of remotely sensed imagery in a tensor affect spatial accuracy? How should models that make spatial predictions be evaluated and compared? I wish I had the answers, but I simply do not know... and am crossing my fingers that smarter folks have a look at this sooner rather than later.

Mach9, a San Francisco-based automated geospatial data production AI platform for infrastructure providers, raised $12 million in seed funding. Quiet Capital led the round and was joined by Kyle Vogt, Amar Hanspal, Scott Belsky, and Gokul Rajaram. https://lnkd.in/edDbRBFT #funding #geospatial #data #ai

This approach provides insight on advancing our understanding of plant fields, which can be further advanced to create 3D interpretable visualizations from complex satellite data. By aligning multi-band satellite imagery with field-level coordinate systems like Mercator or Pseudo-Mercator, parameterized in meters, we can achieve a highly localized analysis that captures variation across the field. AI-driven high-resolution yield and moisture predictions trained from combines allows us to generate 3D volumetric images of the fields leveraging latent representations from neural networks, color-coded by yield, moisture, or moisture-adjusted yield intensity. These 3D visualizations offer agronomists clear, actionable insights into field variability, whether for sales strategies or technical seed product development. For breeders, this approach enhances understanding of how different field trials are related and respond to environmental conditions, aiding in more precise genetic testing. Working with software engineers, we can develop APIs and intuitive GUI tools to make these insights accessible. This incremental approach also sets the stage for groundbreaking innovations, such as VR field simulations that model genetic responses to stress factors like drought or heat. By embracing this technology, we can propel AI-driven agricultural solutions into a new era. #AI #Agriculture #NextGreenRevolution

🌍 Unlocking the Power of Satellite Imagery with PCA: Sentinel-2 Example In satellite imagery analysis, Principle Component Analysis (PCA) can helpe us get more meaningful insights from high-dimensional data. One simple example can be PCA applied to Sentinel-2 multispectral optical imagery. 🔎 What is PCA? PCA is a statistical technique used to reduce the dimensionality of large datasets while preserving the most significant information. With satellite imagery, PCA transforms multiple spectral bands (like those from Sentinel-2) into fewer principal components (PCs): combinations of the original bands that capture the maximum variance in the data. 🛰️ How is PCA Applied to Sentinel-2 Imagery? Sentinel-2 captures data across 13 spectral bands, each highlighting different properties of the Earth's surface. By applying PCA: 1. The data from these bands is consolidated, reducing redundancy between bands. 2. The principal components rank based on the variance they explain, with the first 3 components often carrying the majority of the useful information. 3. By applying first 3 PCs to RGB bands of raster image, this type of visualisation is born: bright magenta areas likely indicate water bodies, while the various shades of green and blue capture different types of vegetation and land cover. PCA can enhances our ability to differentiate between features by condensing the multi-band information into a few composite layers and ,along with other data sources, helpe paint a clearer picture. 🚀 Why Use PCA in Remote Sensing? Dimensionality Reduction Noise Reduction Feature Enhancement #RemoteSensing #Sentinel2 #PCA #MultispectralAnalysis #EarthObservation #Geospatial #DataScience #Sustainability #MONITOREDAI Contains imagery provided by Copernicus Sentinel-2 Created with MONITORED AI - platform developed by OPT/NET BV

🛰️ Revolutionizing Satellite Image Analysis with Computer Vision! 🌍 We’re leveraging the power of Object Detection from Satellite Images using advanced computer vision techniques with Roboflow! 🚀 This cutting-edge technology is enabling faster, more accurate detection of critical objects like infrastructure, land changes, and environmental patterns from space. 🔍 🌟 Key Highlights: Real-time satellite image processing 🕒 Custom object detection models with Roboflow 🔧 Applications in agriculture, urban planning, and disaster management 🌱🏙️ Harnessing satellite data has never been more impactful! Stay tuned as we unlock new insights from above. 🚁 #Pyresearch #AI #ComputerVision #SatelliteImagery #ObjectDetection #Roboflow #SpaceTech #UrbanPlanning #Agriculture #Innovation #DeepLearning #GeospatialAI https://lnkd.in/dDTRzDic

We encourage you to read the recently published article "Enhancing Context Models for Point Cloud Geometry Compression With Context Feature Residuals and Multi-Loss". Authored by: Chang Sun; Hui Yuan; Shuai Li; Xin Lu; Hamzaoui Raouf 📝 Volume: Volume: 14, Issue: 2, June 2024 In point cloud geometry compression, the cross-entropy loss that context models used fails to accurately measure the difference between the one-hot encoding and the predicted probability distribution, and the differences between contexts of different nodes are not significant. To address these problems, we add a multi-layer perception branch, that uses the mean squared error loss function to provide accurate gradients and introduce the context feature residuals into the context model to amplify the differences between contexts. We validate our method by showing that it can improve the performance of an octree- based model (OctAttention) and a voxel-based model (VoxelDNN) on the object point cloud datasets MPEG 8i and MVUB, as well as the LiDAR point cloud dataset SemanticKITTI. 🔗 Continue reading on IEEE Xplore here: https://loom.ly/S8Xevoo #JETCAS #CircuitsandSystems #graphicalabstract #ReadMore

This research paper presents a deep learning-based framework for road marking extraction, classification and modeling from three-dimensional #3D mobile laser scanning (MLS) point clouds. https://lnkd.in/gnAVVzmz

𝐇𝐨𝐰 𝐆𝐞𝐨𝐬𝐩𝐚𝐭𝐢𝐚𝐥 𝟐.𝟎 𝐢𝐬 𝐑𝐞𝐯𝐨𝐥𝐮𝐭𝐢𝐨𝐧𝐢𝐳𝐢𝐧𝐠 𝐃𝐢𝐠𝐢𝐭𝐚𝐥 𝐓𝐰𝐢𝐧𝐬 𝐢𝐧 𝐀𝐄𝐂 The recent Bentley-Cesium merger & the Esri-Autodesk partnership have the potential to revolutionize the AEC industry. But, the importance of Geospatial 2.0 in this new digital twin era cannot be overstated. In this short video, I connect the traditional AEC world to Geospatial 2.0. From infrastructure design to building digital twins to geospatial context: visualization, immersion and AI/Generative AI. 🔗 Stay ahead of Geospatial 2.0—subscribe to our Geospatial 2.0 Insights Newsletter: https://lnkd.in/gGdAS47d #digitaltwin #geospatial #gis #geospatial2point0 #aec #engineering Cesium Bentley Systems