Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

✨ Presenting MicKey (#CVPR2024, Oral) ✨ We regress and match 3D camera coordinates rather then 2D key points, all in metric space. Gives you a scaled relative pose between two RGB images. Paper: Page:

eric_brachmann's profile picture

Eric Brachmann

4,381 subscribers

70,428 Aufrufe • vor 2 Jahren •via X (Twitter)

KunstWissenschaft & TechnologieBildung#CVPR2024
LIVE

Anya Rossi• Live Now

Private livecam show

11 Kommentare

Profilbild von Eric Brachmann
Eric Brachmannvor 2 Jahren

Beyond providing metric estimates, MicKey can also deal with extreme view point changes, up to opposing shots. Here are a few examples of MicKey correspondences.

Profilbild von Eric Brachmann
Eric Brachmannvor 2 Jahren

What is a good key point if you want to go beyond matching co-visible areas? We do not know. MicKey is trained without ground-truth matches, and even without information about whether training image pairs have visual overlap or not.

Profilbild von Eric Brachmann
Eric Brachmannvor 2 Jahren

MicKey is trained using only the relative pose between two images, and differentiable RANSAC (NG-DSAC, #ICCV2021). That means MicKey is trained to establish any correspondence it likes, as long as it leads to a good relative pose.

Profilbild von Eric Brachmann
Eric Brachmannvor 2 Jahren

MicKey does extremely well on the #MapFreeReloc leaderboard, currently ranking second. It's best among the #CVPR2024 cohort, outperforming DUSt3R and FAR. (the field moves fast, and there is already a newer, unknown method that claims top spot)

Profilbild von Eric Brachmann
Eric Brachmannvor 2 Jahren

Code available: Work by @NianticLabs, spearheaded by @axelbarrosotw, with @scriptide and @viprad.

Profilbild von Dan Goldman
Dan Goldmanvor 2 Jahren

Very cool work! IIUC, the scale ambiguity is not really gone, it's just inferring scale from image cues, with a training bias toward open scenes captured by adult human hands... What happens if you use it with images with adversarial cues, like a dollhouse, or trompe l'oeil?

Profilbild von Eric Brachmann
Eric Brachmannvor 2 Jahren

Correct. Given nothing but two RGB images we can only rely on heuristics - and those can be misleading in some cases. Our initial motivation for the #MapFreeReloc benchmark was to see how far heuristics can get us. And after ~1 year I would say: Progress has been astounding.

Profilbild von Dmytro Mishkin 🇺🇦
Dmytro Mishkin 🇺🇦vor 2 Jahren

Well deserved, congrats! P.S. I was faster to post :P

Profilbild von Eric Brachmann
Eric Brachmannvor 2 Jahren

Thanks for sharing, and so fast too! I was trying to post early this morning, but Twitter did not like the video - without telling me why (apsect ratio...)

Profilbild von Dominik Kloepfer
Dominik Kloepfervor 2 Jahren

Pretty cool! If I understand correctly, the 3D keypoints are not actually constrained to lie on any surfaces — how well does the network nonetheless learn to reconstruct pointmaps? Does e.g. depth estimation emerge “for free”?

Profilbild von Eric Brachmann
Eric Brachmannvor 2 Jahren

You are correct. There are no constraints, but still we see reasonable depth maps emerging. One additional twist here is that we only need accurate depth where key points are found. Standard depth estimators try to be accurate everywhere.

Ähnliche Videos

whats interesting is you can "trick" Atlas into becoming a 3D modeling tool by placing input images in 3D space the model then creates a coherent world this is how this scene was made – all the model sees is camera pose + RGB
0:24

whats interesting is you can "trick" Atlas into becoming a 3D modeling tool by placing input images in 3D space the model then creates a coherent world this is how this scene was made – all the model sees is camera pose + RGB

neilson

58,017 Aufrufe • vor 1 Monat

Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors paper page: present Magic123, a two-stage coarse-to-fine approach for high-quality, textured 3D meshes generation from a single unposed image in the wild using both2D and 3D priors. In the first stage, we optimize a neural radiance field to produce a coarse geometry. In the second stage, we adopt a memory-efficient differentiable mesh representation to yield a high-resolution mesh with a visually appealing texture. In both stages, the 3D content is learned through reference view supervision and novel views guided by a combination of 2D and 3D diffusion priors. We introduce a single trade-off parameter between the 2D and 3D priors to control exploration (more imaginative) and exploitation (more precise) of the generated geometry. Additionally, we employ textual inversion and monocular depth regularization to encourage consistent appearances across views and to prevent degenerate solutions, respectively. Magic123 demonstrates a significant improvement over previous image-to-3D techniques, as validated through extensive experiments on synthetic benchmarks and diverse real-world images.
0:34

Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors paper page: present Magic123, a two-stage coarse-to-fine approach for high-quality, textured 3D meshes generation from a single unposed image in the wild using both2D and 3D priors. In the first stage, we optimize a neural radiance field to produce a coarse geometry. In the second stage, we adopt a memory-efficient differentiable mesh representation to yield a high-resolution mesh with a visually appealing texture. In both stages, the 3D content is learned through reference view supervision and novel views guided by a combination of 2D and 3D diffusion priors. We introduce a single trade-off parameter between the 2D and 3D priors to control exploration (more imaginative) and exploitation (more precise) of the generated geometry. Additionally, we employ textual inversion and monocular depth regularization to encourage consistent appearances across views and to prevent degenerate solutions, respectively. Magic123 demonstrates a significant improvement over previous image-to-3D techniques, as validated through extensive experiments on synthetic benchmarks and diverse real-world images.

AK

305,729 Aufrufe • vor 3 Jahren

2D is what the camera sees.👀 3D is what the world is. 🌍 Today, we’re proudly open-sourcing QuerySplat — an open-source feed-forward 3D model that turns a few unposed images into a navigable 3D scene in seconds (run on 4090🪶). Your input: a few unposed images Our output: a sharp, navigable 3D scene — generated in seconds! 🚀 🧠 organize scenes with 3D queries, not pixel-bound Gaussians 📷 no camera poses or manual calibration required ⚡ reconstruct new scenes in a single forward pass ✨ preserve sharp geometry and high-frequency appearance 🏆 achieve state-of-the-art results on DL3DV-Evaluation 🔓 code and model weights are now open-source From pixels to scenes, we are making 3D generation scalable. Try our free iOS app — it’s a lot of fun! 😆 🔗 Try it on the Web: 🔗 Paper: 🔗 Code & weights: 🔗 Project page: 🔗 #AIGC #3DGS #AI #Computervision #3D #artist #3dartist #3d
1:13

2D is what the camera sees.👀 3D is what the world is. 🌍 Today, we’re proudly open-sourcing QuerySplat — an open-source feed-forward 3D model that turns a few unposed images into a navigable 3D scene in seconds (run on 4090🪶). Your input: a few unposed images Our output: a sharp, navigable 3D scene — generated in seconds! 🚀 🧠 organize scenes with 3D queries, not pixel-bound Gaussians 📷 no camera poses or manual calibration required ⚡ reconstruct new scenes in a single forward pass ✨ preserve sharp geometry and high-frequency appearance 🏆 achieve state-of-the-art results on DL3DV-Evaluation 🔓 code and model weights are now open-source From pixels to scenes, we are making 3D generation scalable. Try our free iOS app — it’s a lot of fun! 😆 🔗 Try it on the Web: 🔗 Paper: 🔗 Code & weights: 🔗 Project page: 🔗 #AIGC #3DGS #AI #Computervision #3D #artist #3dartist #3d

InSpatio｜影溯

29,806 Aufrufe • vor 2 Monaten

✨#Warudo New Feature: Camera Background Input 2D 背景をカスタマイズしよう！ 🎇Now you can freely add 2D background to the Warudo interface! This includes custom Images, Videos, Spout sources, and NDI sources. This setting allows the background to remain in a fixed position, unaffected by the control of any other assets. Simply select them in the Camera Asset > Basic Properties > Background Input 😊 #3D #VRM #VTuber
0:39

✨#Warudo New Feature: Camera Background Input 2D 背景をカスタマイズしよう！ 🎇Now you can freely add 2D background to the Warudo interface! This includes custom Images, Videos, Spout sources, and NDI sources. This setting allows the background to remain in a fixed position, unaffected by the control of any other assets. Simply select them in the Camera Asset > Basic Properties > Background Input 😊 #3D #VRM #VTuber

Warudo @ Hakuya Labs

12,520 Aufrufe • vor 4 Tagen

I imported 7 of Jonas’ photos into Agisoft Metashape, where I used its photogrammetric triangulation feature to match key points across the photos. By calculating corresponding points in multiple images, the software determined the camera’s orientation and position in three-dimensional space for each photo. The resulting positions, angles, and distances match what would be expected from a camera on a plane and match the specific details of Jonas’ photos, e.g., zoom, angle, and time between photos. This further proves that the Jonas photos are real, naturally taken photos that have not been manipulated. Their use in the MH370 “satellite” video is extremely strong evidence that the “satellite” video is fake. #MH370x
1:15

I imported 7 of Jonas’ photos into Agisoft Metashape, where I used its photogrammetric triangulation feature to match key points across the photos. By calculating corresponding points in multiple images, the software determined the camera’s orientation and position in three-dimensional space for each photo. The resulting positions, angles, and distances match what would be expected from a camera on a plane and match the specific details of Jonas’ photos, e.g., zoom, angle, and time between photos. This further proves that the Jonas photos are real, naturally taken photos that have not been manipulated. Their use in the MH370 “satellite” video is extremely strong evidence that the “satellite” video is fake. #MH370x

Tony Adams

77,342 Aufrufe • vor 1 Jahr

Let me explain what I'm doing here for those who are interested. I've been building a retro-style D3D12 game engine from scratch, and recently implemented a "Spatial Compression" effect. Essentially its a technique that makes distant geometry appear larger than standard perspective projection would, giving scenes a telephoto compression feel without changing the camera FOV globally (to keep things "normal" when close to the camera). The goal: foreground geometry completely unchanged, objects beyond a configurable start distance progressively scaled up, with a smooth power-curve falloff. The solution: Once the full view-projection transform has been applied, I'm holding 2D screen coordinates. Camera orientation is already resolved at that point, so it's done its job. Scaling those coordinates outward from screen center is a pure 2D operation with no dependency on where the camera is pointing. The scale factor itself is derived from Euclidean world-space distance between the camera and the vertex (not a view-space value, which would rotate with the camera). Combined with the 2D application, both ingredients are fully rotation-invariant. Critically, only the XY screen coordinates are modified. Depth values are left completely untouched, so the depth buffer and any shadow calculations remain correct. I then tried inverting the effect for the opposite in interiors, leading to "Spatial Expansion".
0:29

Let me explain what I'm doing here for those who are interested. I've been building a retro-style D3D12 game engine from scratch, and recently implemented a "Spatial Compression" effect. Essentially its a technique that makes distant geometry appear larger than standard perspective projection would, giving scenes a telephoto compression feel without changing the camera FOV globally (to keep things "normal" when close to the camera). The goal: foreground geometry completely unchanged, objects beyond a configurable start distance progressively scaled up, with a smooth power-curve falloff. The solution: Once the full view-projection transform has been applied, I'm holding 2D screen coordinates. Camera orientation is already resolved at that point, so it's done its job. Scaling those coordinates outward from screen center is a pure 2D operation with no dependency on where the camera is pointing. The scale factor itself is derived from Euclidean world-space distance between the camera and the vertex (not a view-space value, which would rotate with the camera). Combined with the 2D application, both ingredients are fully rotation-invariant. Critically, only the XY screen coordinates are modified. Depth values are left completely untouched, so the depth buffer and any shadow calculations remain correct. I then tried inverting the effect for the opposite in interiors, leading to "Spatial Expansion".

Analog Dream Dev

47,246 Aufrufe • vor 3 Monaten

We are introducing Hunyuan3D-Part: an open-source part-level 3D shape generation model that outperforms all existing open and close-source models. Highlights: 🔹P3-SAM: The industry's first native 3D part segmentation model. 🔹X-Part: A part generation model that achieves state-of-the-art results in controllability and shape quality. Key-features: 1️⃣Eliminates the use of 2D SAM during training, relying solely on a large-scale dataset with 3.7 million shapes and clean part annotations. 2️⃣Introduces a new automated segmentation pipeline in 3D without user intervention. 3️⃣Implements a diffusion-based part decomposition pipeline utilizing both geometry and semantic clues. Code: Weights: Tech reports: 🔸P3-SAM： → Paper: → Project page: 🔸X-Part： → Paper: → Project page: Try it now： → (Light version) Hugging Face demo: → (Full version) Hunyuan3D Studio:
0:39

We are introducing Hunyuan3D-Part: an open-source part-level 3D shape generation model that outperforms all existing open and close-source models. Highlights: 🔹P3-SAM: The industry's first native 3D part segmentation model. 🔹X-Part: A part generation model that achieves state-of-the-art results in controllability and shape quality. Key-features: 1️⃣Eliminates the use of 2D SAM during training, relying solely on a large-scale dataset with 3.7 million shapes and clean part annotations. 2️⃣Introduces a new automated segmentation pipeline in 3D without user intervention. 3️⃣Implements a diffusion-based part decomposition pipeline utilizing both geometry and semantic clues. Code: Weights: Tech reports: 🔸P3-SAM： → Paper: → Project page: 🔸X-Part： → Paper: → Project page: Try it now： → (Light version) Hugging Face demo: → (Full version) Hunyuan3D Studio:

Tencent Hy

72,640 Aufrufe • vor 1 Jahr

✨ We finally crossed a milestone I've been looking forward to! Ever make a sprite and wonder if it actually 'makes sense' in 3D space based on how it looks in 2D? Now we can test that theory! 🦴 Bones animate in real time 🎬 Motion is easier to evaluate 🔧 Fixes are intuitive to notice What's the first tool you built that completely changed how you developed your game? #gamedev #indiedev #animation #pixelart
0:27

✨ We finally crossed a milestone I've been looking forward to! Ever make a sprite and wonder if it actually 'makes sense' in 3D space based on how it looks in 2D? Now we can test that theory! 🦴 Bones animate in real time 🎬 Motion is easier to evaluate 🔧 Fixes are intuitive to notice What's the first tool you built that completely changed how you developed your game? #gamedev #indiedev #animation #pixelart

Walter | MDO Studio

16,863 Aufrufe • vor 2 Monaten

ELON MUSK: “The way to expand compute is really in space. There’s a lot of room in space. And if you look at the size of Earth relative to the Sun or relative to the Solar System, you realize just how tiny it is. We’re very, very tiny. We only receive about half a billionth of the Sun’s energy. So you can really think of Earth as being like a tiny dust mote in a vast darkness. So the way to expand compute without, you know, using up all the land on Earth is to do so in space. And then you can do it without using space for power and water on Earth; you can just do it in space. So I think we’ll probably be launching our first AI satellites next year, and then we’ll be able to do that, I think, at large scale in about two years.”
1:08

ELON MUSK: “The way to expand compute is really in space. There’s a lot of room in space. And if you look at the size of Earth relative to the Sun or relative to the Solar System, you realize just how tiny it is. We’re very, very tiny. We only receive about half a billionth of the Sun’s energy. So you can really think of Earth as being like a tiny dust mote in a vast darkness. So the way to expand compute without, you know, using up all the land on Earth is to do so in space. And then you can do it without using space for power and water on Earth; you can just do it in space. So I think we’ll probably be launching our first AI satellites next year, and then we’ll be able to do that, I think, at large scale in about two years.”

DogeDesigner

65,536 Aufrufe • vor 2 Monaten

Today, we are open-sourcing Hunyuan World 1.1 (WorldMirror), a universal feed-forward 3D reconstruction model. 🚀🚀🚀 While our previously released Hunyuan World 1.0 (open-sourced, lite version deployable on consumer GPUs) focused on generating 3D worlds from text or single-view images, Hunyuan World 1.1 significantly expands the input scope by unlocking video-to-3D and multi-view-to-3D world creation. Highlights: 🔹Any Input, Maximized Flexibility and Fidelity: Flexibly integrates diverse geometric priors (camera poses, intrinsics, depth maps) to resolve structural ambiguities and ensure geometrically consistent 3D outputs. 🔹Any Output, SOTA Results：This elegant architecture simultaneously generates multiple 3D representations: dense point clouds, multi-view depth maps, camera parameters, surface normals, and 3D Gaussian Splattings. 🔹Single-GPU & Fast Inference: As an all-in-one, feed-forward model, Hunyuan World 1.1 runs on a single GPU and delivers all 3D attributes in a single forward pass, within seconds. 🌐Project Page: 🔗Github： 🤗Hugging Face： ✨Demo: 📄Technical Report:
1:20

Today, we are open-sourcing Hunyuan World 1.1 (WorldMirror), a universal feed-forward 3D reconstruction model. 🚀🚀🚀 While our previously released Hunyuan World 1.0 (open-sourced, lite version deployable on consumer GPUs) focused on generating 3D worlds from text or single-view images, Hunyuan World 1.1 significantly expands the input scope by unlocking video-to-3D and multi-view-to-3D world creation. Highlights: 🔹Any Input, Maximized Flexibility and Fidelity: Flexibly integrates diverse geometric priors (camera poses, intrinsics, depth maps) to resolve structural ambiguities and ensure geometrically consistent 3D outputs. 🔹Any Output, SOTA Results：This elegant architecture simultaneously generates multiple 3D representations: dense point clouds, multi-view depth maps, camera parameters, surface normals, and 3D Gaussian Splattings. 🔹Single-GPU & Fast Inference: As an all-in-one, feed-forward model, Hunyuan World 1.1 runs on a single GPU and delivers all 3D attributes in a single forward pass, within seconds. 🌐Project Page: 🔗Github： 🤗Hugging Face： ✨Demo: 📄Technical Report:

Tencent Hy

168,848 Aufrufe • vor 11 Monaten

Check out our #PAMI paper with code "Dense Continuous-Time Optical Flow from Event Cameras," where we show how to regress *continuous-time* trajectories of every pixel from event cameras alone or events plus frames! The key idea is to iteratively estimate per-pixel polynomials using a recurrent lookup and update scheme. Paper: Code: DOI: We present a method for estimating dense continuous-time optical flow from event data. Traditional dense optical flow methods compute the pixel displacement between two images. Due to missing information, these approaches cannot recover the pixel trajectories in the blind time between two images. We show that it is possible to compute per-pixel, continuous-time optical flow using events from an event camera. Events provide temporally fine-grained information about movement in pixel space due to their asynchronous nature and microsecond response time. We leverage these benefits to predict pixel trajectories densely in continuous time via parameterized Bézier curves. To achieve this, we build a neural network with strong inductive biases for this task: First, we build multiple sequential correlation volumes in time using event data. Second, we use Bézier curves to index these correlation volumes at multiple timestamps along the trajectory. Third, we use the retrieved correlation to update the Bézier curve representations iteratively. Our method can optionally include image pairs to boost performance further. To train and evaluate our model, we introduce a synthetic dataset (MultiFlow) that features moving objects and ground truth trajectories for every pixel. Our quantitative experiments suggest that our method successfully predicts pixel trajectories in continuous time and is competitive in the traditional two-view pixel displacement metric on MultiFlow and DSEC-Flow. Open source code and datasets are released to the public. Kudos to Mathias Gehrig Manasi Muglikar
0:23

Check out our #PAMI paper with code "Dense Continuous-Time Optical Flow from Event Cameras," where we show how to regress *continuous-time* trajectories of every pixel from event cameras alone or events plus frames! The key idea is to iteratively estimate per-pixel polynomials using a recurrent lookup and update scheme. Paper: Code: DOI: We present a method for estimating dense continuous-time optical flow from event data. Traditional dense optical flow methods compute the pixel displacement between two images. Due to missing information, these approaches cannot recover the pixel trajectories in the blind time between two images. We show that it is possible to compute per-pixel, continuous-time optical flow using events from an event camera. Events provide temporally fine-grained information about movement in pixel space due to their asynchronous nature and microsecond response time. We leverage these benefits to predict pixel trajectories densely in continuous time via parameterized Bézier curves. To achieve this, we build a neural network with strong inductive biases for this task: First, we build multiple sequential correlation volumes in time using event data. Second, we use Bézier curves to index these correlation volumes at multiple timestamps along the trajectory. Third, we use the retrieved correlation to update the Bézier curve representations iteratively. Our method can optionally include image pairs to boost performance further. To train and evaluate our model, we introduce a synthetic dataset (MultiFlow) that features moving objects and ground truth trajectories for every pixel. Our quantitative experiments suggest that our method successfully predicts pixel trajectories in continuous time and is competitive in the traditional two-view pixel displacement metric on MultiFlow and DSEC-Flow. Open source code and datasets are released to the public. Kudos to Mathias Gehrig Manasi Muglikar

Davide Scaramuzza

12,692 Aufrufe • vor 2 Jahren

🚨 Paper Alert 🚨 ➡️Paper Title: Articulate3D: Zero-Shot Text-Driven 3D Object Posing 🌟Few pointers from the paper 🎯Authors of this paper proposed a training-free method, “Articulate3D”, to pose a 3D asset through language control. 🎯Despite advances in vision and language models, this task remains surprisingly challenging. 🎯To achieve this goal, they decomposed the problem into two steps. 🎯They modified a powerful image-generator to create target images conditioned on the input image and a text instruction. 🎯They then align the mesh to the target images through a multi-view pose optimisation step. 🎯 In detail, they introduced a self-attention rewiring mechanism (RSActrl) that decouples the source structure from pose within an image generative model, allowing it to maintain a consistent structure across varying poses. 🎯They observed that differentiable rendering is an unreliable signal for articulation optimisation; instead, they used keypoints to establish correspondences between input and target images. 🎯The effectiveness of Articulate3D is demonstrated across a diverse range of 3D objects and free-form text prompts, successfully manipulating poses while maintaining the original identity of the mesh. 🎯Quantitative evaluations and a comparative user study, in which their method was preferred over 85% of the time, confirm its superiority over existing approaches. 🏢Organization: University of Oxford , Google DeepMind 🧙Paper Authors: Oishi Deb, Anjun Hu, Ashkan Khakzar, Philip Torr, Christian Rupprecht 📝 Read the Full Paper here: 🗂️ Project Page: 🎥 Be sure to watch the attached Demo Video - Sound on 🔊🔊 Find this Valuable 💎 ? ♻️QT and teach your network something new Follow me 👣, naveen manwani , for the latest updates on Tech and AI-related news, insightful research papers, and exciting announcements.
0:46

🚨 Paper Alert 🚨 ➡️Paper Title: Articulate3D: Zero-Shot Text-Driven 3D Object Posing 🌟Few pointers from the paper 🎯Authors of this paper proposed a training-free method, “Articulate3D”, to pose a 3D asset through language control. 🎯Despite advances in vision and language models, this task remains surprisingly challenging. 🎯To achieve this goal, they decomposed the problem into two steps. 🎯They modified a powerful image-generator to create target images conditioned on the input image and a text instruction. 🎯They then align the mesh to the target images through a multi-view pose optimisation step. 🎯 In detail, they introduced a self-attention rewiring mechanism (RSActrl) that decouples the source structure from pose within an image generative model, allowing it to maintain a consistent structure across varying poses. 🎯They observed that differentiable rendering is an unreliable signal for articulation optimisation; instead, they used keypoints to establish correspondences between input and target images. 🎯The effectiveness of Articulate3D is demonstrated across a diverse range of 3D objects and free-form text prompts, successfully manipulating poses while maintaining the original identity of the mesh. 🎯Quantitative evaluations and a comparative user study, in which their method was preferred over 85% of the time, confirm its superiority over existing approaches. 🏢Organization: University of Oxford , Google DeepMind 🧙Paper Authors: Oishi Deb, Anjun Hu, Ashkan Khakzar, Philip Torr, Christian Rupprecht 📝 Read the Full Paper here: 🗂️ Project Page: 🎥 Be sure to watch the attached Demo Video - Sound on 🔊🔊 Find this Valuable 💎 ? ♻️QT and teach your network something new Follow me 👣, naveen manwani , for the latest updates on Tech and AI-related news, insightful research papers, and exciting announcements.

naveen manwani

14,334 Aufrufe • vor 1 Jahr

Want to use Depth Anything, but need metric depth rather than relative depth? Thrilled to introduce Prompt Depth Anything, a new paradigm for accurate metric depth estimation with up to 4K resolution. 👉Key Message: Depth foundation models like DA have already internalized rich geometric knowledge of the 3D world but lack a proper way to elicit it. Inspired by the success of prompting in LLMs, we propose prompting Depth Anything with metric cues to produce metric depth. This method proves to be very effective when using a low-cost lidar (e.g., iPhone's LiDAR), which is widely available, as prompts. We believe the prompt can generalize to other forms as long as scale information is provided. Prompt Depth Anything offers 1⃣A series of models for iPhone lidars. 2⃣4D reconstruction from monocular videos (captured with iPhone). 3⃣Improved generalization ability for robot manipulation, e.g. Training on cans but generalizing on glasses. 4⃣More detailed depth annotations for the ScanNet++ dataset. The first author is our excellent intern Haotong Lin. Paper: Huggingface: Project Page: Code:
1:20

Want to use Depth Anything, but need metric depth rather than relative depth? Thrilled to introduce Prompt Depth Anything, a new paradigm for accurate metric depth estimation with up to 4K resolution. 👉Key Message: Depth foundation models like DA have already internalized rich geometric knowledge of the 3D world but lack a proper way to elicit it. Inspired by the success of prompting in LLMs, we propose prompting Depth Anything with metric cues to produce metric depth. This method proves to be very effective when using a low-cost lidar (e.g., iPhone's LiDAR), which is widely available, as prompts. We believe the prompt can generalize to other forms as long as scale information is provided. Prompt Depth Anything offers 1⃣A series of models for iPhone lidars. 2⃣4D reconstruction from monocular videos (captured with iPhone). 3⃣Improved generalization ability for robot manipulation, e.g. Training on cans but generalizing on glasses. 4⃣More detailed depth annotations for the ScanNet++ dataset. The first author is our excellent intern Haotong Lin. Paper: Huggingface: Project Page: Code:

Bingyi Kang

67,750 Aufrufe • vor 1 Jahr

3D-LLM: Injecting the 3D World into Large Language Models paper page: Large language models (LLMs) and Vision-Language Models (VLMs) have been proven to excel at multiple tasks, such as commonsense reasoning. Powerful as these models can be, they are not grounded in the 3D physical world, which involves richer concepts such as spatial relationships, affordances, physics, layout, and so on. In this work, we propose to inject the 3D world into large language models and introduce a whole new family of 3D-LLMs. Specifically, 3D-LLMs can take 3D point clouds and their features as input and perform a diverse set of 3D-related tasks, including captioning, dense captioning, 3D question answering, task decomposition, 3D grounding, 3D-assisted dialog, navigation, and so on. Using three types of prompting mechanisms that we design, we are able to collect over 300k 3D-language data covering these tasks. To efficiently train 3D-LLMs, we first utilize a 3D feature extractor that obtains 3D features from rendered multi- view images. Then, we use 2D VLMs as our backbones to train our 3D-LLMs. By introducing a 3D localization mechanism, 3D-LLMs can better capture 3D spatial information. Experiments on ScanQA show that our model outperforms state-of-the-art baselines by a large margin (e.g., the BLEU-1 score surpasses state-of-the-art score by 9%). Furthermore, experiments on our held-in datasets for 3D captioning, task composition, and 3D-assisted dialogue show that our model outperforms 2D VLMs. Qualitative examples also show that our model could perform more tasks beyond the scope of existing LLMs and VLMs.
0:50

3D-LLM: Injecting the 3D World into Large Language Models paper page: Large language models (LLMs) and Vision-Language Models (VLMs) have been proven to excel at multiple tasks, such as commonsense reasoning. Powerful as these models can be, they are not grounded in the 3D physical world, which involves richer concepts such as spatial relationships, affordances, physics, layout, and so on. In this work, we propose to inject the 3D world into large language models and introduce a whole new family of 3D-LLMs. Specifically, 3D-LLMs can take 3D point clouds and their features as input and perform a diverse set of 3D-related tasks, including captioning, dense captioning, 3D question answering, task decomposition, 3D grounding, 3D-assisted dialog, navigation, and so on. Using three types of prompting mechanisms that we design, we are able to collect over 300k 3D-language data covering these tasks. To efficiently train 3D-LLMs, we first utilize a 3D feature extractor that obtains 3D features from rendered multi- view images. Then, we use 2D VLMs as our backbones to train our 3D-LLMs. By introducing a 3D localization mechanism, 3D-LLMs can better capture 3D spatial information. Experiments on ScanQA show that our model outperforms state-of-the-art baselines by a large margin (e.g., the BLEU-1 score surpasses state-of-the-art score by 9%). Furthermore, experiments on our held-in datasets for 3D captioning, task composition, and 3D-assisted dialogue show that our model outperforms 2D VLMs. Qualitative examples also show that our model could perform more tasks beyond the scope of existing LLMs and VLMs.

AK

249,798 Aufrufe • vor 3 Jahren

After a year of team work, we're thrilled to introduce Depth Anything 3 (DA3)! 🚀 Aiming for human-like spatial perception, DA3 extends monocular depth estimation to any-view scenarios, including single images, multi-view images, and video. In pursuit of minimal modeling, DA3 reveals two key insights: 💎 A plain transformer (e.g., vanilla DINO) is enough. No specialized architecture. ✨ A single depth-ray representation is enough. No complex 3D tasks. Three series of models have been released: the main DA3 series, a monocular metric estimation series, and a monocular depth estimation series. The core team members, aside from me: Haotong Lin, Sili Chen, Jun Hao Liew, Donny Y. Chen. 👇(1/n) #DepthAnything3
0:32

After a year of team work, we're thrilled to introduce Depth Anything 3 (DA3)! 🚀 Aiming for human-like spatial perception, DA3 extends monocular depth estimation to any-view scenarios, including single images, multi-view images, and video. In pursuit of minimal modeling, DA3 reveals two key insights: 💎 A plain transformer (e.g., vanilla DINO) is enough. No specialized architecture. ✨ A single depth-ray representation is enough. No complex 3D tasks. Three series of models have been released: the main DA3 series, a monocular metric estimation series, and a monocular depth estimation series. The core team members, aside from me: Haotong Lin, Sili Chen, Jun Hao Liew, Donny Y. Chen. 👇(1/n) #DepthAnything3

Bingyi Kang

516,195 Aufrufe • vor 10 Monaten

Code and data are now online for CameraHMR, our state-of-the-art parametric 3D human pose and shape (HPS) estimation method that will appear at hashtag#3DV2025. There are 4 key contributions that make it so accurate and robust: 1. To get accurate 3D shape and pose as well as good alignment to image features, you need to know the focal length of the camera. To solve this, we train HumanFOV to compute the field of view. 2. We introduce CameraHMR, which integrates HumanFOV into HMR2.0 to exploit the estimated focal length. 3. To get accurate pseudo ground truth (pGT) training data, we compute the focal length for images in 4DHumans dataset and modify SMPLify to take this into account. 4. But SMPLify only uses sparse 2D keypoints, which do not capture body shape. So we train a dense surface keypoint detector, DenseKP, on BEDLAM and run it on 4DHumans, resulting in improved body shape. The resulting method is CamSMPLify. We iterate training CameraHMR and running CamSMPLify on the training set initialized with CameraHMR. This results in much improved pGT for 4DHumans and a SOTA single-image HMR method.
4:38

Code and data are now online for CameraHMR, our state-of-the-art parametric 3D human pose and shape (HPS) estimation method that will appear at hashtag#3DV2025. There are 4 key contributions that make it so accurate and robust: 1. To get accurate 3D shape and pose as well as good alignment to image features, you need to know the focal length of the camera. To solve this, we train HumanFOV to compute the field of view. 2. We introduce CameraHMR, which integrates HumanFOV into HMR2.0 to exploit the estimated focal length. 3. To get accurate pseudo ground truth (pGT) training data, we compute the focal length for images in 4DHumans dataset and modify SMPLify to take this into account. 4. But SMPLify only uses sparse 2D keypoints, which do not capture body shape. So we train a dense surface keypoint detector, DenseKP, on BEDLAM and run it on 4DHumans, resulting in improved body shape. The resulting method is CamSMPLify. We iterate training CameraHMR and running CamSMPLify on the training set initialized with CameraHMR. This results in much improved pGT for 4DHumans and a SOTA single-image HMR method.

Michael Black

21,696 Aufrufe • vor 1 Jahr

Mistral AI Releases Robostral Navigate: An 8B Model Enabling Robots to Navigate Complex Environments Hitting 76.6% on R2R-CE With One RGB Camera. No LiDAR. No depth sensor. No multi-camera rig. Here's how it works. 👇 1. Pointing, not metric commands The model predicts the pixel coordinates of the next target in the camera view, plus the arrival orientation. Working in pixel space keeps it robust to camera intrinsics and world scale. When the target leaves the frame, it falls back to local displacements ("2m forward, 1.5m left, turn 25°"). 2. Grounding-first No open-source VLM base. It starts from Mistral's grounding model (pointing, counting, localization). Navigation emerges once the model knows where things are. → ~400,000 trajectories across 6,000 simulated scenes 3. Prefix-caching for training A tree-based attention mask packs a full episode into one sequence — all time steps in a single forward pass. → 22× fewer training tokens; months of training done in days 4. Online RL on top After supervised training, CISPO adds trial-and-error learning to fight distribution shift from behavior cloning. → +3.2% success rate from RL alone 5. The numbers (R2R-CE, Matterport3D) → 76.6% success on validation unseen → +9.7 pts over best single-camera approach → +4.5 pts over best depth/multi-camera system The key takeaway: state-of-the-art continuous VLN without a sensor stack — grounding-init, pixel-space actions, prefix-cached SFT, and online RL, on one RGB camera. Full analysis: Technical details: Mistral AI Mistral AI for Developers
0:22

Mistral AI Releases Robostral Navigate: An 8B Model Enabling Robots to Navigate Complex Environments Hitting 76.6% on R2R-CE With One RGB Camera. No LiDAR. No depth sensor. No multi-camera rig. Here's how it works. 👇 1. Pointing, not metric commands The model predicts the pixel coordinates of the next target in the camera view, plus the arrival orientation. Working in pixel space keeps it robust to camera intrinsics and world scale. When the target leaves the frame, it falls back to local displacements ("2m forward, 1.5m left, turn 25°"). 2. Grounding-first No open-source VLM base. It starts from Mistral's grounding model (pointing, counting, localization). Navigation emerges once the model knows where things are. → ~400,000 trajectories across 6,000 simulated scenes 3. Prefix-caching for training A tree-based attention mask packs a full episode into one sequence — all time steps in a single forward pass. → 22× fewer training tokens; months of training done in days 4. Online RL on top After supervised training, CISPO adds trial-and-error learning to fight distribution shift from behavior cloning. → +3.2% success rate from RL alone 5. The numbers (R2R-CE, Matterport3D) → 76.6% success on validation unseen → +9.7 pts over best single-camera approach → +4.5 pts over best depth/multi-camera system The key takeaway: state-of-the-art continuous VLN without a sensor stack — grounding-init, pixel-space actions, prefix-cached SFT, and online RL, on one RGB camera. Full analysis: Technical details: Mistral AI Mistral AI for Developers

Marktechpost AI

39,955 Aufrufe • vor 2 Monaten

SceNeRFlow: Time-Consistent Reconstruction of General Dynamic Scenes abs: paper page: Existing methods for the 4D reconstruction of general, non-rigidly deforming objects focus on novel-view synthesis and neglect correspondences. However, time consistency enables advanced downstream tasks like 3D editing, motion analysis, or virtual-asset creation. We propose SceNeRFlow to reconstruct a general, non-rigid scene in a time-consistent manner. Our dynamic-NeRF method takes multi-view RGB videos and background images from static cameras with known camera parameters as input. It then reconstructs the deformations of an estimated canonical model of the geometry and appearance in an online fashion. Since this canonical model is time-invariant, we obtain correspondences even for long-term, long-range motions. We employ neural scene representations to parametrize the components of our method. Like prior dynamic-NeRF methods, we use a backwards deformation model. We find non-trivial adaptations of this model necessary to handle larger motions: We decompose the deformations into a strongly regularized coarse component and a weakly regularized fine component, where the coarse component also extends the deformation field into the space surrounding the object, which enables tracking over time. We show experimentally that, unlike prior work that only handles small motion, our method enables the reconstruction of studio-scale motions.
0:26

SceNeRFlow: Time-Consistent Reconstruction of General Dynamic Scenes abs: paper page: Existing methods for the 4D reconstruction of general, non-rigidly deforming objects focus on novel-view synthesis and neglect correspondences. However, time consistency enables advanced downstream tasks like 3D editing, motion analysis, or virtual-asset creation. We propose SceNeRFlow to reconstruct a general, non-rigid scene in a time-consistent manner. Our dynamic-NeRF method takes multi-view RGB videos and background images from static cameras with known camera parameters as input. It then reconstructs the deformations of an estimated canonical model of the geometry and appearance in an online fashion. Since this canonical model is time-invariant, we obtain correspondences even for long-term, long-range motions. We employ neural scene representations to parametrize the components of our method. Like prior dynamic-NeRF methods, we use a backwards deformation model. We find non-trivial adaptations of this model necessary to handle larger motions: We decompose the deformations into a strongly regularized coarse component and a weakly regularized fine component, where the coarse component also extends the deformation field into the space surrounding the object, which enables tracking over time. We show experimentally that, unlike prior work that only handles small motion, our method enables the reconstruction of studio-scale motions.

AK

76,380 Aufrufe • vor 3 Jahren

✨ I can now generate 3d assets for my drone sim at directly from Cursor (sponsor of #vibejam) I need buildings that you'd see in a war torn city, like warehouses in ruins, broken down abandoned houses, bombed out bridges etc. Nano Banana Pro or 2 can generate them really well and then you can put them in an image-to-3d model and you get a GLB or FBX That one you can then import into your Three.js game, the models might be big though, in my case like 16MB, so I ask it to compress it and make it more low poly so it loads fast ThreeJS then loads the individual GLBs on page load and puts them in my drone sim somewhere randomly, I think I should remove some of the grass and match the sandy color of the ruins though to make it fit in more
0:20

✨ I can now generate 3d assets for my drone sim at directly from Cursor (sponsor of #vibejam) I need buildings that you'd see in a war torn city, like warehouses in ruins, broken down abandoned houses, bombed out bridges etc. Nano Banana Pro or 2 can generate them really well and then you can put them in an image-to-3d model and you get a GLB or FBX That one you can then import into your Three.js game, the models might be big though, in my case like 16MB, so I ask it to compress it and make it more low poly so it loads fast ThreeJS then loads the individual GLBs on page load and puts them in my drone sim somewhere randomly, I think I should remove some of the grass and match the sandy color of the ruins though to make it fit in more

@levelsio

134,777 Aufrufe • vor 5 Monaten