New tutorial | Monocular depth estimation with Depth Anything v3 🌊 Learn how to generate depth maps from a single RGB image or video frame and explore real-world applications in robotics, navigation, & AR/VR. Watch here ➡️ bit.ly/4vqCGO7 #Ultralytics #DepthEstimation #ComputerVision

🚨 2D fotoğraflardan **metrik olarak doğru 3D veri** çıkarmak artık çok daha basit ve hassas hâle geldi! 🔥 Princeton Vision & Learning Lab’tan yeni bir yöntem: **ZEDD** Focus stack (farklı odak noktalarındaki birden fazla fotoğraf) kullanarak 2D görüntülerden 3D derinlik bilgisi üretiyor. ### ZEDD’in öne çıkan üstünlükleri: • DepthPro’ya göre **U daha yüksek doğruluk** (ZEDD benchmark’ında) • Sadece **2 kare** ile çalışabiliyor • ViT-B backbone DPT decoder mimarisi • Özel donanım veya lidar’a gerek yok Bu gelişme, özellikle mobil cihazlar, drone’lar, AR/VR ve robotik uygulamalar için büyük bir sıçrama anlamına geliyor. 🔗 GitHub: github.com/princeton-vl/FOSS… Sizce ZEDD, 3D modelleme ve derinlik algılama dünyasını nasıl değiştirecek? Mobil fotoğrafçılıkta mı, robotikte mi, yoksa AR uygulamalarında mı en büyük etkiyi yaratacak? Yorumlarda düşüncelerinizi paylaşın, en yaratıcı kullanım alanlarını birlikte keşfedelim! 📸➡️3️⃣ #ZEDD #3DReconstruction #FocusStack #DepthEstimation #PrincetonVL #AI #ComputerVision #YapayZeka #3DModeling #ARVR

Very excited to share our first public release after I joined @robbyant_brain! We present Lingbot-Depth 👀 — a state-of-the-art depth foundation model trained with RGB-D MAE on millions of real & simulated RGBD pairs. 🔹 Camera depths as natural masks for RGB-D MAE modeling 🔹 Large-scale real sim RGBD data curation pipeline What surprised us most: ✨ Significant improvement on transparent, reflective surfaces, and thin structures — traditionally the hardest cases for depth models. 👇These are some of hard causal-capture images we tested on. The results are pretty good! #DepthEstimation #FoundationModels #EmbodiedAI #3DVision #Robotics #RobbyAnt

Takeaway: Masked Depth Modeling turns missing depth into supervision, producing strong depth priors that generalize across perception and robotics tasks. Model: huggingface.co/robbyant/ling… Code: github.com/Robbyant/lingbot-… • • Tech Report: github.com/Robbyant/lingbot-… #CVPR2025 #DepthEstimation #RGBD #Robotics

New tutorial | Depth estimation with MiDaS by Intel Labs See how AI perceives distance with monocular depth, run demos, visualize results & perform real-time webcam inference. Watch here ➡️ ow.ly/krmY50Xn35m #DepthEstimation #MachineLearning

📢 Happening now! @gene_ch0u, @wenqi_xian, @stevenygd, and @realNingYu are presenting FlashDepth: Real-time Streaming Video Depth Estimation at 2K Resolution, at #ICCV2025. [Poster] 🗓️ Today (Tue, Oct 21) | 🕑 3-5pm HST | 📍 Hawaii Convention Center ExHall II Poster 433 👋 To continue our research momentum, Eyeline Labs is hiring research interns — flexible in time, location, and focus. We publish and open-source, aiming to raise the bar together on next-gen AI filmmaking tools. Fall’25/Spring’26: jobs.lever.co/Eyeline/4eded8… Summer’26: jobs.lever.co/Eyeline/85f377… #AI #AIResearch #AIGC #GenAI #ComputerVision #MachineLearning #DepthEstimation #Internship #hiring

🎉 Congrats on DAC and the CVPR’25 acceptance! We explored a similar idea for stereo in FoVA-Depth (3DV’24 oral): 📌 Warp pinhole to canonical ERP to generalize across camera models. Glad to see this working in monocular too! 📄 Project: research.nvidia.com/labs/lpr… 💻 Code: github.com/NVlabs/fova-depth 🧵 Original tweet: x.com/daniel_lichy/status/18… #CVPR2025 #DepthEstimation #360Images

Fine-tuned DepthAnythingV2 using real-world ROVR data — in collaboration with leading global startups and research labs. 📷 Left: original image 📊 Right: depth prediction after fine-tuning 📈 Huge accuracy boost: d1: 0.2348 → 0.6894 abs_rel: 0.9753 → 0.2793 log10: ∞ → 0.2263 silog: NaN → 1.0851 ✅ Real 3D data changes everything. 🌍 This is just the beginning — ROVR is building the world’s largest open 3D dataset. More AI breakthroughs coming soon. #ROVR #LiDAR #AI #DepthEstimation #DePIN #ComputerVision #OpenSourceAI

Introducing AIRPHA by AquaAge – a drone platform that uses depth estimation for autonomous indoor navigation. No LiDAR needed. Just AI and smart perception. youtube.com/watch?v=AfAWn8qB… #DroneTech #AIRPHA #IndoorFlight #DepthEstimation #AI #AquaAge

Efficiency–Accuracy Trade-Off in Light Field Estimation with Cost Volume Construction and Aggregation mdpi.com/1424-8220/24/11/358… #depthestimation #convolutionneuralnetwork

📢"Prompting Depth Anything for 4K Resolution Accurate Metric Depth Estimation" Key Highlights: ✅Foundation Model LiDAR Prompts: Introduces a prompt fusion framework leveraging low-cost LiDAR data to guide a Depth Anything foundation model, achieving metric-accurate depth estimation at 4K resolution. ✅Multi-Scale Prompt Fusion: Integrates LiDAR signals at multiple scales within the decoder, enhancing depth estimation across diverse scenes and resolutions. ✅Scalable Data Pipeline: Proposes a synthetic LiDAR simulation and pseudo ground truth generation strategy to compensate for the lack of high-quality real-world paired datasets. ✅SOTA Performance: Outperforms existing monocular depth estimation approaches on ARKitScenes and ScanNet , setting new benchmarks for accuracy and generalization. ✅Real-World Applications: Demonstrates impact on 3D reconstruction, robotic grasping, and AR/VR applications, ensuring scale-consistent metric depth for downstream tasks. --- Paper: arxiv.org/abs/2412.14015 Project Page: promptda.github.io/ Github: github.com/DepthAnything/Pro… ---- Related articles from LearnOpenCV: 1. Metric Depth with Depth Pro: learnopencv.com/depth-pro-mo… 2. Monocular Depth Estimation - Depth Anything: learnopencv.com/depth-anythi… 3. Disparity Estimation using Deep Learning: learnopencv.com/disparity-es… ---- #MonocularDepth #DepthEstimation #DepthAnything #ByteDance #FoundationalModel

Expanding Sparse Radar Depth Based on Joint Bilateral Filter for Radar-Guided Monocular Depth Estimation mdpi.com/1424-8220/24/6/1864 #depthestimation #multimodality #radar

🌟Breaking News! 🎉 Excited to announce that our paper, “A Unified Monocular Vision-Based Driving Model for Autonomous Vehicles with Multi-Task Capabilities,” has been accepted in IEEE Transactions on Intelligent Vehicles (T-IV)! 🚗👀 @salimazak , @ffbozkaya, Ş. Tığlıoğlu, A. Yusefi, and A. Durdu 📄 DOI: 10.1109/TIV.2024.3483114 📊 Journal Impact Factor (2023): 14.0 🏆 Top Journal Rankings: #ComputerScience, #ArtificialIntelligence: 5/197 #Engineering, #ElectricalandElectronic: 5/353 #TransportationScienceTechnology: 2/72 @KtunRacLab🙏 URL: ieeexplore.ieee.org/document… #AutonomousDriving #SelfDriving #ComputerVision #Monocular #SensorFusion #IEEE #LSTM #DepthEstimation #Transactions #IntelligentVehicles

"DEPTH PRO: SHARP MONOCULAR METRIC DEPTH IN LESS THAN A SECOND" @alexeyab84 Apple introduces Depth Pro, a revolutionary model for zero-shot metric monocular depth estimation, setting new standards in speed, precision, and versatility. This research paves the way for highly efficient and accurate depth estimation across various real-world applications. Key Highlights: ✅ High-resolution depth maps: Depth Pro produces exceptionally sharp 2.25-megapixel depth maps with detailed boundary tracing, even for challenging features like hair and fine structures. ✅ Incredibly fast performance: With a processing time of less than 0.3 seconds, Depth Pro is perfect for real-time applications, delivering fast, high-quality results. ✅ Zero-shot adaptability: It works without requiring any camera metadata, producing metric depth maps with absolute scale for any image, making it highly adaptable across different scenarios. ✅ Multi-scale Vision Transformer architecture: Depth Pro uses an advanced ViT-based design to balance global image context with fine-grained details, ensuring both accuracy and efficiency. ✅ Improved boundary tracing: By using real-world and synthetic datasets during training, Depth Pro excels in tracing boundaries and maintaining sharpness, significantly reducing visual artifacts like “flying pixels.” Compared to models like Marigold and DepthAnything v2, Depth Pro delivers faster and sharper depth maps. While Marigold produces detailed boundaries, it is far slower in performance. Depth Pro not only offers superior sharpness in boundary precision but also achieves this at a fraction of the time, making it ideal for interactive applications like 3D photography and novel view synthesis.Depth Pro redefines what's possible in depth estimation, opening new doors for real-time image editing and rendering. Project Page: arxiv.org/abs/2410.02073 Paper: arxiv.org/abs/2410.02073 Github: github.com/apple/ml-depth-pr… #DepthEstimation #AIResearch #VisionTransformers #ComputerVision #AppleResearch #3DPhotography #MonocularDepth #DeepLearning #RealTimeDepth #NovelViewSynthesis #ImageProcessing

New video is up about metric depth estimation combined with the new YOLO11 model from @ultralytics In this video we are going to take a look at how to estimate real distances to objects with just a single camera or image and YOLO11. We are going to use the new depth pro model from apple which is used for metric depth estimation. We then combine the depth map with object detections from YOLO11. In just a few lines of code you can estimate the distance to objects. #yolo11 #objectdetection #depth #depthestimation #metricdepth #ai

💬𝐌𝐞𝐭𝐚 𝐛𝐮𝐢𝐥𝐝𝐬 𝐭𝐡𝐞 𝐛𝐫𝐢𝐝𝐠𝐞 𝐭𝐨 𝐦𝐞𝐭𝐚𝐯𝐞𝐫𝐬𝐞 🚀 Cr. Meta via Ahsen Khaliq #ai #foundationHumanModel #meta #segmentation #depthEstimation

4/25 Foundation Model for Robotic Endoscopic Surgery This paper presents Depth Anything in Robotic Endoscopic Surgery (DARES), which introduces Vector-LoRA, a new adaptation technique for self-supervised monocular depth estimation in robotic-assisted surgery (RAS). #SurgicalRobotics #DepthEstimation #AI x.com/OpenlifesciAI/status/1…

👁 How can the field of robotics leverage different eye apertures in nature?👁 🤖 As my first post as a faculty at the University of Colorado Boulder, we introduce #CodedVO: Coded Visual Odometry. (Published in #IEEE #RAL Journal) 🌟 Key Highlights: 🥇 A novel monocular visual odometry method 🥈 Leverages optical constraints from coded apertures to resolve scale ambiguity. 🥉 By integrating RGB and predicted metric depth using optical constraints, we achieve state-of-the-art performance in monocular visual odometry with a known scale in indoor scenes. #Robotics #VisualOdometry #ComputerVision #DepthEstimation #ComputationalImaging #UMD #ICRA40 #IEEE #RAL

🚀 Announcing CodedVO: A Leap in Monocular Visual Odometry! 🌟 📰 Excited to share our research, "Coded Visual Odometry: CodedVO," which has been accepted in IEEE RAL Journal and will be presented at ICRA40! #Robotics #VisualOdometry #ComputerVision #DepthEstimation #ComputationalImaging #UMD #ICRA40 #IEEE #RAL

Monday’s research paper This study tackles the challenge of estimating depth using monocular vision for moving objects in outdoor scenes using a self-supervised approach. By introducing a new training strategy that leverages the ground contact prior, the authors enhance depth estimation accuracy. Their method shows significant improvements on Cityscapes and KITTI datasets, particularly for moving objects. Authors: Jaeho Moon, Juan Luis Gonzalez Bello, Byeongjun Kwon, and Munchurl Kim. Paper: arxiv.org/abs/2312.10118 Project Page: kaist-viclab.github.io/From_… #AI #MachineLearning #ComputerVision #DepthEstimation #SelfSupervisedLearning #Research #Cityscapes #KITTI #OpenCV #OpenCVUniversity #MondayResearch #AIResearch