Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

LLaVA-3D A Simple yet Effective Pathway to Empowering LMMs with 3D-awareness Recent advancements in Large Multimodal Models (LMMs) have greatly enhanced their proficiency in 2D visual understanding tasks, enabling them to effectively process and understand images and videos. However, the development of LMMs with 3D-awareness for 3D scene understanding... has been hindered by the lack of large-scale 3D vision-language datasets and powerful 3D encoders. In this paper, we introduce a simple yet effective framework called LLaVA-3D. Leveraging the strong 2D understanding priors from LLaVA, our LLaVA-3D efficiently adapts LLaVA for 3D scene understanding without compromising 2D understanding capabilities. To achieve this, we employ a simple yet effective representation, 3D Patch, which connects 2D CLIP patch features with their corresponding positions in 3D space. By integrating the 3D Patches into 2D LMMs and employing joint 2D and 3D vision-language instruction tuning, we establish a unified architecture for both 2D image understanding and 3D scene understanding. Experimental results show that LLaVA-3D converges 3.5x faster than existing 3D LMMs when trained on 3D vision-language datasets. Moreover, LLaVA-3D not only achieves state-of-the-art performance across various 3D tasks but also maintains comparable 2D image understanding and vision-language conversation capabilities with LLaVA.show more

AK

502,568 subscribers

41,713 views • 1 year ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

2 Comments

AK1 year ago

discuss:

Yang1 year ago

Build powerful AI agents and automations for your business today. Check out our step-by-step video tutorials 100% FREE 🥳

Related Videos

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.

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

249,572 views • 2 years ago

Introducing SAM 3D, the newest addition to the SAM collection, bringing common sense 3D understanding of everyday images. SAM 3D includes two models: 🛋️ SAM 3D Objects for object and scene reconstruction 🧑‍🤝‍🧑 SAM 3D Body for human pose and shape estimation Both models achieve state-of-the-art performance transforming static 2D images into vivid, accurate reconstructions. 🔗 Learn more:

Introducing SAM 3D, the newest addition to the SAM collection, bringing common sense 3D understanding of everyday images. SAM 3D includes two models: 🛋️ SAM 3D Objects for object and scene reconstruction 🧑‍🤝‍🧑 SAM 3D Body for human pose and shape estimation Both models achieve state-of-the-art performance transforming static 2D images into vivid, accurate reconstructions. 🔗 Learn more:

AI at Meta

858,011 views • 7 months ago

3D-VLA A 3D Vision-Language-Action Generative World Model Recent vision-language-action (VLA) models rely on 2D inputs, lacking integration with the broader realm of the 3D physical world. Furthermore, they perform action prediction by learning a direct mapping from

3D-VLA A 3D Vision-Language-Action Generative World Model Recent vision-language-action (VLA) models rely on 2D inputs, lacking integration with the broader realm of the 3D physical world. Furthermore, they perform action prediction by learning a direct mapping from

AK

84,940 views • 2 years ago

3D-R1 is out on Hugging Face Enhancing Reasoning in 3D VLMs for Unified Scene Understanding 3D-R1 is an open-source generalist model that enhances the reasoning of 3D VLMs for unified scene understanding

3D-R1 is out on Hugging Face Enhancing Reasoning in 3D VLMs for Unified Scene Understanding 3D-R1 is an open-source generalist model that enhances the reasoning of 3D VLMs for unified scene understanding

AK

34,729 views • 10 months ago

The challenge of creating this scene was from a 3D environment to a 2D painted environment. We used the character to bait-and-switch out the backgrounds. #2d #3d #animation

The challenge of creating this scene was from a 3D environment to a 2D painted environment. We used the character to bait-and-switch out the backgrounds. #2d #3d #animation

THE LINE

696,502 views • 3 years ago

Combining 2D and 3D in After Effects. In Foldspace, not every surface needs to be bent. Keep one flat for 2D artwork, and let the others exist in 3D space. This allows you to combine 2D and 3D objects in a single animation with no masks or tricks needed.

Combining 2D and 3D in After Effects. In Foldspace, not every surface needs to be bent. Keep one flat for 2D artwork, and let the others exist in 3D space. This allows you to combine 2D and 3D objects in a single animation with no masks or tricks needed.

Motion Princess

21,666 views • 3 months ago

3D Animator Nicolas Roth and Digital Artist Boris have combined 2D and 3D techniques to achieve a truly unique art style in their latest mixed media project. Breakdown of the project: #3danimation #2danimation #2d #3d #3dcg #vfx #b3d #blender #mixedmedia

3D Animator Nicolas Roth and Digital Artist Boris have combined 2D and 3D techniques to achieve a truly unique art style in their latest mixed media project. Breakdown of the project: #3danimation #2danimation #2d #3d #3dcg #vfx #b3d #blender #mixedmedia

80 LEVEL

57,988 views • 1 year ago

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.

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

305,643 views • 3 years ago

Semantically annotating 3D gaussian splats on the fly using gemini 3.1 + sparkjs 1. Load any 3D scene and hit scan 2. Get 2D detections from VLM 3. Cluster outputs & project into 3D world space 4. Save as a persistent 3D semantic layer Inspired by Alexander Chen's experiments with gemini visual intelligence. Just had to try to lift it from 2D to 3D!

Semantically annotating 3D gaussian splats on the fly using gemini 3.1 + sparkjs 1. Load any 3D scene and hit scan 2. Get 2D detections from VLM 3. Cluster outputs & project into 3D world space 4. Save as a persistent 3D semantic layer Inspired by Alexander Chen's experiments with gemini visual intelligence. Just had to try to lift it from 2D to 3D!

Bilawal Sidhu

56,568 views • 1 month ago

The past decade of AI was about understanding 2D data. The next is likely all about 3D and 4D structures.

The past decade of AI was about understanding 2D data. The next is likely all about 3D and 4D structures.

a16z

62,707 views • 1 year ago

📢WorldAgents: 3D worlds only from 2D image models - without any training! We propose an agentic approach with a Director (VLM) to plan the scene, a Generator (Flux or NanoBanana) for new views, and a Verifier (VLM) for selection / 3D consistency. -> High-fidelity 3D worlds from a single text prompt. What's remarkable: our agents find consistent views from 2D image models to obtain 3D-consistent worlds; this shows that image models contain world priors - agents just need to find them! Great work by Ziya Erkoç Angela Dai

📢WorldAgents: 3D worlds only from 2D image models - without any training! We propose an agentic approach with a Director (VLM) to plan the scene, a Generator (Flux or NanoBanana) for new views, and a Verifier (VLM) for selection / 3D consistency. -> High-fidelity 3D worlds from a single text prompt. What's remarkable: our agents find consistent views from 2D image models to obtain 3D-consistent worlds; this shows that image models contain world priors - agents just need to find them! Great work by Ziya Erkoç Angela Dai

Matthias Niessner

18,886 views • 3 months ago

Introducing 𝐒𝐀𝐑𝟑𝐃, which tokenizes 3D objects into multiscale tokens and generates 3D objects by autoregressive next-scale prediction. 𝐒𝐀𝐑𝟑𝐃 enables fast 3D generation and comprehensive 3D understanding. arXiv: Project:

Introducing 𝐒𝐀𝐑𝟑𝐃, which tokenizes 3D objects into multiscale tokens and generates 3D objects by autoregressive next-scale prediction. 𝐒𝐀𝐑𝟑𝐃 enables fast 3D generation and comprehensive 3D understanding. arXiv: Project:

Xingang Pan

28,248 views • 1 year ago

⚡🎉 We are thrilled to introduce VORTEX, an AI-powered computational framework for predicting 3D Spatial Transcriptomics (ST) using 3D tissue images and minimal 2D ST! 🧬 By combining cutting-edge 3D non-destructive tissue imaging with AI, VORTEX imputes the 3D molecular landscape of large tissue samples in a cost-effective and scalable manner. 🧠💡Our approach: By pretraining on diverse 3D morphology–2D transcriptomic pairs from heterogeneous tissue samples, and then fine-tuning on minimal 2D ST data from a volume of interest, VORTEX leverages both generic tissue-specific and sample-specific morphomolecular correlates to predict 3D ST. Congratulations to our superstar co-leads Cristina Almagro Pérez and Andrew H. Song, this was an exciting collaboration with Jonathan Liu Sizun Jiang Ali Bashashati. Preprint: Demo: Read the excellent blog from our superstar grad student Cristina Almagro Pérez: Also see our previous work on 3D Computational Pathology from Andrew H. Song published in Cell last year: Stay tuned for more to come.

⚡🎉 We are thrilled to introduce VORTEX, an AI-powered computational framework for predicting 3D Spatial Transcriptomics (ST) using 3D tissue images and minimal 2D ST! 🧬 By combining cutting-edge 3D non-destructive tissue imaging with AI, VORTEX imputes the 3D molecular landscape of large tissue samples in a cost-effective and scalable manner. 🧠💡Our approach: By pretraining on diverse 3D morphology–2D transcriptomic pairs from heterogeneous tissue samples, and then fine-tuning on minimal 2D ST data from a volume of interest, VORTEX leverages both generic tissue-specific and sample-specific morphomolecular correlates to predict 3D ST. Congratulations to our superstar co-leads Cristina Almagro Pérez and Andrew H. Song, this was an exciting collaboration with Jonathan Liu Sizun Jiang Ali Bashashati. Preprint: Demo: Read the excellent blog from our superstar grad student Cristina Almagro Pérez: Also see our previous work on 3D Computational Pathology from Andrew H. Song published in Cell last year: Stay tuned for more to come.

Faisal Mahmood

17,991 views • 1 year ago

Can we synthesize 3D human-scene interactions without learning from any 3D data? Yes! Check out Lei Li's GenZI, a novel zero-shot approach to generating 3D interactions by distilling priors from large vision-language models.

Can we synthesize 3D human-scene interactions without learning from any 3D data? Yes! Check out Lei Li's GenZI, a novel zero-shot approach to generating 3D interactions by distilling priors from large vision-language models.

Angela Dai

106,862 views • 2 years ago

📢 OneCanvas: 3D Scene Understanding via Panoramic Reprojection We extract features from video frames and reproject them into one occlusion-free view of the whole scene that a 2D VLM reads just like a normal image. We can center this view on any viewpoint, including an agent's own pose for situated reasoning. The same projection lets us create spatial training tasks with no human annotation, solvable only by reasoning over the 3D positions of real object features placed on an otherwise empty canvas. The result is a stock 2D VLM that reasons in 3D, setting a new state of the art across spatial benchmarks at far less compute. 🌐 ▶️ Great work by Bartłomiej Baranowski & Dave Zhenyu Chen

📢 OneCanvas: 3D Scene Understanding via Panoramic Reprojection We extract features from video frames and reproject them into one occlusion-free view of the whole scene that a 2D VLM reads just like a normal image. We can center this view on any viewpoint, including an agent's own pose for situated reasoning. The same projection lets us create spatial training tasks with no human annotation, solvable only by reasoning over the 3D positions of real object features placed on an otherwise empty canvas. The result is a stock 2D VLM that reasons in 3D, setting a new state of the art across spatial benchmarks at far less compute. 🌐 ▶️ Great work by Bartłomiej Baranowski & Dave Zhenyu Chen

Matthias Niessner

24,276 views • 8 days ago

3D Artist Rito_Art showcased this vibrant fox character that combines 2D and 3D features. Details about the tools and more artwork:

3D Artist Rito_Art showcased this vibrant fox character that combines 2D and 3D features. Details about the tools and more artwork:

80 LEVEL

27,994 views • 4 months ago

DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion Prior paper page: present DreamCraft3D, a hierarchical 3D content generation method that produces high-fidelity and coherent 3D objects. We tackle the problem by leveraging a 2D reference image to guide the stages of geometry sculpting and texture boosting. A central focus of this work is to address the consistency issue that existing works encounter. To sculpt geometries that render coherently, we perform score distillation sampling via a view-dependent diffusion model. This 3D prior, alongside several training strategies, prioritizes the geometry consistency but compromises the texture fidelity. We further propose Bootstrapped Score Distillation to specifically boost the texture. We train a personalized diffusion model, Dreambooth, on the augmented renderings of the scene, imbuing it with 3D knowledge of the scene being optimized. The score distillation from this 3D-aware diffusion prior provides view-consistent guidance for the scene. Notably, through an alternating optimization of the diffusion prior and 3D scene representation, we achieve mutually reinforcing improvements: the optimized 3D scene aids in training the scene-specific diffusion model, which offers increasingly view-consistent guidance for 3D optimization. The optimization is thus bootstrapped and leads to substantial texture boosting. With tailored 3D priors throughout the hierarchical generation, DreamCraft3D generates coherent 3D objects with photorealistic renderings, advancing the state-of-the-art in 3D content generation.

DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion Prior paper page: present DreamCraft3D, a hierarchical 3D content generation method that produces high-fidelity and coherent 3D objects. We tackle the problem by leveraging a 2D reference image to guide the stages of geometry sculpting and texture boosting. A central focus of this work is to address the consistency issue that existing works encounter. To sculpt geometries that render coherently, we perform score distillation sampling via a view-dependent diffusion model. This 3D prior, alongside several training strategies, prioritizes the geometry consistency but compromises the texture fidelity. We further propose Bootstrapped Score Distillation to specifically boost the texture. We train a personalized diffusion model, Dreambooth, on the augmented renderings of the scene, imbuing it with 3D knowledge of the scene being optimized. The score distillation from this 3D-aware diffusion prior provides view-consistent guidance for the scene. Notably, through an alternating optimization of the diffusion prior and 3D scene representation, we achieve mutually reinforcing improvements: the optimized 3D scene aids in training the scene-specific diffusion model, which offers increasingly view-consistent guidance for 3D optimization. The optimization is thus bootstrapped and leads to substantial texture boosting. With tailored 3D priors throughout the hierarchical generation, DreamCraft3D generates coherent 3D objects with photorealistic renderings, advancing the state-of-the-art in 3D content generation.

AK

161,530 views • 2 years ago

Multi-modal #LLMs understand a lot about humans. But do they understand our 3D pose? We train #PoseGPT to estimate, generate, and reason about 3D human pose (#SMPL) in images and text. This is the first true foundation model for understanding 3D humans.

Multi-modal #LLMs understand a lot about humans. But do they understand our 3D pose? We train #PoseGPT to estimate, generate, and reason about 3D human pose (#SMPL) in images and text. This is the first true foundation model for understanding 3D humans.

Michael Black

81,398 views • 2 years ago

Perfect blend of 2d and 3d animation

Perfect blend of 2d and 3d animation

Anime Aesthetics

193,051 views • 1 year ago

Digital Artist Ram showcased a truly jaw-dropping building drawing that blurs the line between 2D and 3D. The artwork was made with Feather 3D, a sketching app that allows artists to essentially "draw in 3D":

Digital Artist Ram showcased a truly jaw-dropping building drawing that blurs the line between 2D and 3D. The artwork was made with Feather 3D, a sketching app that allows artists to essentially "draw in 3D":

80 LEVEL

230,417 views • 1 year ago