Video yükleniyor...

Video Yüklenemedi

Ana Sayfaya Dön

Freshly accepted to #ICCV2023! Trained on **images only**, we can freely extrapolate into or out of a scene. Joint effort with an amazing team Eric Chan Songyou Peng Mohamad Shahbazi Anton Obukhov Luc Van Gool Gordon Wetzstein

prime_cai's profile picture

Prime (Shengqu) Cai

1,904 subscribers

66,412 görüntüleme • 3 yıl önce •via X (Twitter)

#ICCV2023
LIVE

Anya Rossi• Live Now

Private livecam show

0 Yorum

Yorum bulunmuyor

Orijinal gönderinin yorumları burada görünecek

Benzer Videolar

Sharing something exciting we've been working on as a Thanksgiving gift: Diffusion Self-Distillation (DSD), which redefines zero-shot customized image generation using FLUX. DSD is like DreamBooth, but zero-shot/training-free. It works across any input subject and desired context—character consistency, item/asset adaptation, scene relighting, and more. It even enables the creation of comics/mangas without any effort in fine-tuning or training a personalized model! 📰 Paper: 🌐 Website: Team effort with Eric Chan, Yunzhi Zhang, Leonidas Guibas, Jiajun Wu, and Gordon Wetzstein.
0:42

Sharing something exciting we've been working on as a Thanksgiving gift: Diffusion Self-Distillation (DSD), which redefines zero-shot customized image generation using FLUX. DSD is like DreamBooth, but zero-shot/training-free. It works across any input subject and desired context—character consistency, item/asset adaptation, scene relighting, and more. It even enables the creation of comics/mangas without any effort in fine-tuning or training a personalized model! 📰 Paper: 🌐 Website: Team effort with Eric Chan, Yunzhi Zhang, Leonidas Guibas, Jiajun Wu, and Gordon Wetzstein.

Prime (Shengqu) Cai

60,760 görüntüleme • 1 yıl önce

Learning from robot data? Standard. Direct Video-Action Models (DVA) is different: treat robot control as video generation, then translate the generated video into actions. Built by , the system pre-trains causal video models from scratch and can run complex production tasks for hours using only ~10 hours of robot data. • hundreds of frames of visual context • real-time control via causal video prediction More: The team behind it just exited 18 months of stealth with a $450M Series A at a $1.7B valuation. Founded by Jagdeep Singh (ex-QuantumScape) with a Stanford-heavy science team: CSO Eric Ryan Chan (ex-WorldLabs) and Prof. Gordon Wetzstein. Already running in large-scale automotive production environments. Backed by Vinod Khosla Ventures, Temasek, Premji Invest, and John Doerr.​​​​​​​​​​​​​​​​ Thanks for sharing, Tongzhou Mu 🤖🦾🦿 👋
6:01

Learning from robot data? Standard. Direct Video-Action Models (DVA) is different: treat robot control as video generation, then translate the generated video into actions. Built by , the system pre-trains causal video models from scratch and can run complex production tasks for hours using only ~10 hours of robot data. • hundreds of frames of visual context • real-time control via causal video prediction More: The team behind it just exited 18 months of stealth with a $450M Series A at a $1.7B valuation. Founded by Jagdeep Singh (ex-QuantumScape) with a Stanford-heavy science team: CSO Eric Ryan Chan (ex-WorldLabs) and Prof. Gordon Wetzstein. Already running in large-scale automotive production environments. Backed by Vinod Khosla Ventures, Temasek, Premji Invest, and John Doerr.​​​​​​​​​​​​​​​​ Thanks for sharing, Tongzhou Mu 🤖🦾🦿 👋

Ilir Aliu

26,391 görüntüleme • 6 ay önce

The LVM (large vision model) revolution is coming a little after the LLM (large language model) one, and will transform how we process images. But there’s an important difference between LVMs and LLMs: - Internet text is similar enough to proprietary text documents that an LLM trained on internet text can understand your documents. - But internet images – such as Instagram pictures – contain a lot of pictures of people, pets, landmarks, and everyday objects. Many practical vision applications (manufacturing, aerial imagery, life sciences, etc.) use images that look nothing like most internet images. So a generic LVM trained on internet images fares poorly at picking out the most salient features of images in many specialized domains. That’s why domain specific LVMs – ones adapted to images of a particular domain (such as semiconductor manufacturing, or pathology) – do much better. At LandingAI , by using ~100K unlabeled images to adapt an LVM to a specific domain, we see significantly improved results, for example where only 10-30% as much labeled data is now needed to achieve a certain level of performance. For companies with large sets of images that look nothing like internet images, I think domain specific LVMs can be a way to unlock considerable value from their data. Dan Maloney and I share more details in the video.
3:55

The LVM (large vision model) revolution is coming a little after the LLM (large language model) one, and will transform how we process images. But there’s an important difference between LVMs and LLMs: - Internet text is similar enough to proprietary text documents that an LLM trained on internet text can understand your documents. - But internet images – such as Instagram pictures – contain a lot of pictures of people, pets, landmarks, and everyday objects. Many practical vision applications (manufacturing, aerial imagery, life sciences, etc.) use images that look nothing like most internet images. So a generic LVM trained on internet images fares poorly at picking out the most salient features of images in many specialized domains. That’s why domain specific LVMs – ones adapted to images of a particular domain (such as semiconductor manufacturing, or pathology) – do much better. At LandingAI , by using ~100K unlabeled images to adapt an LVM to a specific domain, we see significantly improved results, for example where only 10-30% as much labeled data is now needed to achieve a certain level of performance. For companies with large sets of images that look nothing like internet images, I think domain specific LVMs can be a way to unlock considerable value from their data. Dan Maloney and I share more details in the video.

Andrew Ng

685,054 görüntüleme • 2 yıl önce

Elon Musk explains how we can achieve abundance for all with AI and robotics “With AI and robotics, this is the path to abundance for all If you want to solve global poverty or give everyone a very high standard of living, the only way to do that is AI and robotics" if AI and robotics can handle the production of goods and services at near-zero cost, we could effectively eliminate poverty Elon Musk is going to lead humanity into an age of Amazing Abundance
1:04

Elon Musk explains how we can achieve abundance for all with AI and robotics “With AI and robotics, this is the path to abundance for all If you want to solve global poverty or give everyone a very high standard of living, the only way to do that is AI and robotics" if AI and robotics can handle the production of goods and services at near-zero cost, we could effectively eliminate poverty Elon Musk is going to lead humanity into an age of Amazing Abundance

X Freeze

14,332 görüntüleme • 16 gün önce

Differentiable Blocks World: Qualitative 3D Decomposition by Rendering Primitives paper page: Given a set of calibrated images of a scene, we present an approach that produces a simple, compact, and actionable 3D world representation by means of 3D primitives. While many approaches focus on recovering high-fidelity 3D scenes, we focus on parsing a scene into mid-level 3D representations made of a small set of textured primitives. Such representations are interpretable, easy to manipulate and suited for physics-based simulations. Moreover, unlike existing primitive decomposition methods that rely on 3D input data, our approach operates directly on images through differentiable rendering. Specifically, we model primitives as textured superquadric meshes and optimize their parameters from scratch with an image rendering loss. We highlight the importance of modeling transparency for each primitive, which is critical for optimization and also enables handling varying numbers of primitives. We show that the resulting textured primitives faithfully reconstruct the input images and accurately model the visible 3D points, while providing amodal shape completions of unseen object regions. We compare our approach to the state of the art on diverse scenes from DTU, and demonstrate its robustness on real-life captures from BlendedMVS and Nerfstudio. We also showcase how our results can be used to effortlessly edit a scene or perform physical simulations.
0:27

Differentiable Blocks World: Qualitative 3D Decomposition by Rendering Primitives paper page: Given a set of calibrated images of a scene, we present an approach that produces a simple, compact, and actionable 3D world representation by means of 3D primitives. While many approaches focus on recovering high-fidelity 3D scenes, we focus on parsing a scene into mid-level 3D representations made of a small set of textured primitives. Such representations are interpretable, easy to manipulate and suited for physics-based simulations. Moreover, unlike existing primitive decomposition methods that rely on 3D input data, our approach operates directly on images through differentiable rendering. Specifically, we model primitives as textured superquadric meshes and optimize their parameters from scratch with an image rendering loss. We highlight the importance of modeling transparency for each primitive, which is critical for optimization and also enables handling varying numbers of primitives. We show that the resulting textured primitives faithfully reconstruct the input images and accurately model the visible 3D points, while providing amodal shape completions of unseen object regions. We compare our approach to the state of the art on diverse scenes from DTU, and demonstrate its robustness on real-life captures from BlendedMVS and Nerfstudio. We also showcase how our results can be used to effortlessly edit a scene or perform physical simulations.

AK

38,571 görüntüleme • 3 yıl önce

Today, I held an extended meeting with our foreign affairs team regarding our interaction with NATO – both with the Alliance as a whole and specifically with our neighboring NATO member countries. It is evident that together we can add more security and stability to our region and all of Europe. This evident potential must be translated into practical steps, demonstrating how we can provide each other with real security through cooperation. Today, we discussed our joint capabilities to defend against Russian missile and drone strikes and are actively working on their implementation. I thank everyone who is helping.
3:05

Today, I held an extended meeting with our foreign affairs team regarding our interaction with NATO – both with the Alliance as a whole and specifically with our neighboring NATO member countries. It is evident that together we can add more security and stability to our region and all of Europe. This evident potential must be translated into practical steps, demonstrating how we can provide each other with real security through cooperation. Today, we discussed our joint capabilities to defend against Russian missile and drone strikes and are actively working on their implementation. I thank everyone who is helping.

Volodymyr Zelenskyy / Володимир Зеленський

322,521 görüntüleme • 2 yıl önce

Bryson DeChambeau made the US Ryder Cup team automatically despite playing just 8 qualifying events across the 2 year qualification period. Keegan Bradley acknowledged that achievement in his press conference yesterday: “A lot to say about Bryson. One of the most incredible things is he qualified for this team off of eight starts, which is unheard of. But what I'm most impressed about Bryson is the effort that he's made to be a part of this team. He's had to travel, go out of his way to meet us in our places, and he's gone above and beyond what we would ever ask of a player, and that's the thing I'm most proud of. “You can go on and on about what an incredible player he is and golfer, but what he's done for this team, putting in the extra effort, is amazing to see.” Bryson is going to be a real spectacle at Bethpage in front of the New York fans!
0:34

Bryson DeChambeau made the US Ryder Cup team automatically despite playing just 8 qualifying events across the 2 year qualification period. Keegan Bradley acknowledged that achievement in his press conference yesterday: “A lot to say about Bryson. One of the most incredible things is he qualified for this team off of eight starts, which is unheard of. But what I'm most impressed about Bryson is the effort that he's made to be a part of this team. He's had to travel, go out of his way to meet us in our places, and he's gone above and beyond what we would ever ask of a player, and that's the thing I'm most proud of. “You can go on and on about what an incredible player he is and golfer, but what he's done for this team, putting in the extra effort, is amazing to see.” Bryson is going to be a real spectacle at Bethpage in front of the New York fans!

Flushing It

681,780 görüntüleme • 1 yıl önce

How SEAL Team 6 Used an Empty Helicopter to Lure Out Nearly 30 Taliban “We were following this group of Taliban, about 30 of them. We knew if we went toe-to-toe with them, some of our guys were definitely going to get killed or wounded. As a ruse or a fake-out, we brought in an empty helicopter close to the valley to make a lot of noise. They started to get out and run, and we waited for all of them to get out in the open and ambushed all of them.” “I pull the M67 frag grenade, and I try to get the perfect angle. I drop it in perfectly, and I turn to run. And I slip. I start sliding down into the hole of the bunker with the hand grenade. I could feel it rolling on my shin. The hand grenade with the pin pulled. I started crawling my way out and finally got out of the bunker just in time for it to be like, boom. Almost completely blew myself up.” Alex West | Advocate & Speaker
16:45

How SEAL Team 6 Used an Empty Helicopter to Lure Out Nearly 30 Taliban “We were following this group of Taliban, about 30 of them. We knew if we went toe-to-toe with them, some of our guys were definitely going to get killed or wounded. As a ruse or a fake-out, we brought in an empty helicopter close to the valley to make a lot of noise. They started to get out and run, and we waited for all of them to get out in the open and ambushed all of them.” “I pull the M67 frag grenade, and I try to get the perfect angle. I drop it in perfectly, and I turn to run. And I slip. I start sliding down into the hole of the bunker with the hand grenade. I could feel it rolling on my shin. The hand grenade with the pin pulled. I started crawling my way out and finally got out of the bunker just in time for it to be like, boom. Almost completely blew myself up.” Alex West | Advocate & Speaker

Shawn Ryan

133,608 görüntüleme • 1 ay önce

📢📢 𝐀𝐯𝐚𝐭𝟑𝐫 📢📢 Avat3r creates high-quality 3D head avatars from just a few input images in a single forward pass with a new dynamic 3DGS reconstruction model. Video: Project: Our core idea is to make Gaussian Reconstruction Models animatable. We find that a simple cross-attention to an expression code sequence is already sufficient to model complex facial expressions. We then incorporate position maps from DUSt3R and feature maps from Sapiens to facilitate the prediction task. While DUSt3R's position maps act as a pixel-aligned initialization for the Gaussians' positions, the Sapiens feature maps help the cross-view transformer to match corresponding image tokens in the 4 input images. One major challenge in creating a 3D head avatar from smartphone images comes from inconsistent facial expressions when the subject could not remain perfectly static during the capture. We eliminate this static requirement by simply showing our model input images with different facial expressions during training. This technique makes our model robust to inconsistent input images later on. Finally, we show that despite the model has been trained with 4 input images, one can even create a 3D head avatar when only a single image is available. To achieve this, we employ a pre-trained 3D GAN to lift the single image to 3D and then render the 4 input images for our model. This allows us to create 3D head avatars from single images and even highly out-of-distribution examples like AI generated faces, paintings or statues. Great work by Tobias Kirschstein from his internship at Meta with Javier Romero, Artem Sevastopolsky, and Shunsuke Saito
0:38

📢📢 𝐀𝐯𝐚𝐭𝟑𝐫 📢📢 Avat3r creates high-quality 3D head avatars from just a few input images in a single forward pass with a new dynamic 3DGS reconstruction model. Video: Project: Our core idea is to make Gaussian Reconstruction Models animatable. We find that a simple cross-attention to an expression code sequence is already sufficient to model complex facial expressions. We then incorporate position maps from DUSt3R and feature maps from Sapiens to facilitate the prediction task. While DUSt3R's position maps act as a pixel-aligned initialization for the Gaussians' positions, the Sapiens feature maps help the cross-view transformer to match corresponding image tokens in the 4 input images. One major challenge in creating a 3D head avatar from smartphone images comes from inconsistent facial expressions when the subject could not remain perfectly static during the capture. We eliminate this static requirement by simply showing our model input images with different facial expressions during training. This technique makes our model robust to inconsistent input images later on. Finally, we show that despite the model has been trained with 4 input images, one can even create a 3D head avatar when only a single image is available. To achieve this, we employ a pre-trained 3D GAN to lift the single image to 3D and then render the 4 input images for our model. This allows us to create 3D head avatars from single images and even highly out-of-distribution examples like AI generated faces, paintings or statues. Great work by Tobias Kirschstein from his internship at Meta with Javier Romero, Artem Sevastopolsky, and Shunsuke Saito

Matthias Niessner

74,818 görüntüleme • 1 yıl önce

The Saint Paul Police Department has been using unmanned aircraft systems, or drones, since 2023. This technology provides officers with real-time information from above an active scene. Now in 2026, we’re expanding our drone program to help all first responders. Our team of trained drone pilots can now provide a live thermal view above a fire – allowing the Saint Paul Fire Dept to identify hot spots and develop the best plan of action. There are now two drone options: • A Saint Paul Police officer, who is also a trained and authorized drone pilot, can respond and launch a drone at the scene. • Or we can deploy one of our three new docked drones. Docked drones are remotely piloted, can be on scene in less than 3 minutes, and being to relay real-time information to responding officers or firefighters before they even arrive. These tools are improving public safety, reducing response times, and leading to better-informed decisions and better outcomes for everyone in the City of Saint Paul. • Learn about our Docked Drone Program and other public safety technology tools, plus view our interactive flight dashboard, policies, and more: • Read more via the Pioneer Press:
1:10

The Saint Paul Police Department has been using unmanned aircraft systems, or drones, since 2023. This technology provides officers with real-time information from above an active scene. Now in 2026, we’re expanding our drone program to help all first responders. Our team of trained drone pilots can now provide a live thermal view above a fire – allowing the Saint Paul Fire Dept to identify hot spots and develop the best plan of action. There are now two drone options: • A Saint Paul Police officer, who is also a trained and authorized drone pilot, can respond and launch a drone at the scene. • Or we can deploy one of our three new docked drones. Docked drones are remotely piloted, can be on scene in less than 3 minutes, and being to relay real-time information to responding officers or firefighters before they even arrive. These tools are improving public safety, reducing response times, and leading to better-informed decisions and better outcomes for everyone in the City of Saint Paul. • Learn about our Docked Drone Program and other public safety technology tools, plus view our interactive flight dashboard, policies, and more: • Read more via the Pioneer Press:

Saint Paul Police Department

17,680 görüntüleme • 3 ay önce

A few survival tricks we all can use when out on a hike or camping in the wilderness. First this gentleman shows us how to filter water using a t-shirt to keep bugs and debris out of the water. Next he boiled the water. This is an important art before drinking any water especially from creeks and stagnant water. Finally he shows us how to sharpen an axe or knife using water and gritty dirt on a stone. It’s an amazing lesson. Thankfully, I have a small grinder for such a situation. But not everyone has one with them. I hope this helps someone in the future.
0:53

A few survival tricks we all can use when out on a hike or camping in the wilderness. First this gentleman shows us how to filter water using a t-shirt to keep bugs and debris out of the water. Next he boiled the water. This is an important art before drinking any water especially from creeks and stagnant water. Finally he shows us how to sharpen an axe or knife using water and gritty dirt on a stone. It’s an amazing lesson. Thankfully, I have a small grinder for such a situation. But not everyone has one with them. I hope this helps someone in the future.

🌸𝓐𝓾𝓭𝓻𝓮𝔂🌸

61,147 görüntüleme • 1 ay önce

The one, the only, the amazing Florence Pugh is in the house ladies and gentlemen. I THOUGHT SHE WAS AMERICAN before I prepped the interview, that’s how out of it I am, but also how amazing SHE IS with her accents and her ability to disappear into roles. Midsommar? Little Women? The Wonder? Lady Macbeth? Don’t Worry Darling? Black Widow??? All stand-out performances. We dip into these and other roles and also find out a little bit about the woman behind them. Enjoy! And you’re welcome! You can find the LTP on Spotify.
1:07

The one, the only, the amazing Florence Pugh is in the house ladies and gentlemen. I THOUGHT SHE WAS AMERICAN before I prepped the interview, that’s how out of it I am, but also how amazing SHE IS with her accents and her ability to disappear into roles. Midsommar? Little Women? The Wonder? Lady Macbeth? Don’t Worry Darling? Black Widow??? All stand-out performances. We dip into these and other roles and also find out a little bit about the woman behind them. Enjoy! And you’re welcome! You can find the LTP on Spotify.

Louis Theroux

682,910 görüntüleme • 10 ay önce

We have not properly communicated what we do. I have no interest in causing an environmental disaster on my legacy. That’s never going to happen. We will bring plans out for a one gig facility and show what the heat profile looks like, the light, the noise, the water, all of it. When we build this, you can come and kick the tires and extrapolate from this. We need to become more ingrained with the community and I think that is going to be incorporated into all projects. If I’ve said anything that offends you or done anything that you think has been unfair, I apologize. But I also would say I want you to lend an open ear to what I’m going to show you in the months ahead.
1:41

We have not properly communicated what we do. I have no interest in causing an environmental disaster on my legacy. That’s never going to happen. We will bring plans out for a one gig facility and show what the heat profile looks like, the light, the noise, the water, all of it. When we build this, you can come and kick the tires and extrapolate from this. We need to become more ingrained with the community and I think that is going to be incorporated into all projects. If I’ve said anything that offends you or done anything that you think has been unfair, I apologize. But I also would say I want you to lend an open ear to what I’m going to show you in the months ahead.

Kevin O'Leary aka Mr. Wonderful

41,330 görüntüleme • 1 ay önce

I made this tool 4 years ago for helping with customer support emails. Instead of typing, it allows you to compose an email by dragging from predefined blocks with variables. So you only write your blocks once, and then you only combine them with drag and drop into an email. You can also save a set of email blocks as a template. When you choose that template, it lets you fill the variables of the template like a form, and then send it. We only use it internally and now with all the gpt stuff going on it seems pointless to release it. or is it? 🤔
0:54

I made this tool 4 years ago for helping with customer support emails. Instead of typing, it allows you to compose an email by dragging from predefined blocks with variables. So you only write your blocks once, and then you only combine them with drag and drop into an email. You can also save a set of email blocks as a template. When you choose that template, it lets you fill the variables of the template like a form, and then send it. We only use it internally and now with all the gpt stuff going on it seems pointless to release it. or is it? 🤔

kitze the 🐐

72,424 görüntüleme • 3 yıl önce

The Americas alone are COVERED with proof that an ancient civilization once resided on Earth, which was far technologically-superior to ours. There is simply no other explanation after considering the plethora of evidence that's visibly right in front of us. The only question now is...what do we do with this information? We must start with coming to terms with the fact that ALL of our history textbooks are lies. Throw everything you thought you knew about our past out the window, open your mind, and never look back. We have the greatest puzzle of all time right in front of us, and it's going to take a team effort to solve 🧩🤔 [🎥yt: archivist]
37:38

The Americas alone are COVERED with proof that an ancient civilization once resided on Earth, which was far technologically-superior to ours. There is simply no other explanation after considering the plethora of evidence that's visibly right in front of us. The only question now is...what do we do with this information? We must start with coming to terms with the fact that ALL of our history textbooks are lies. Throw everything you thought you knew about our past out the window, open your mind, and never look back. We have the greatest puzzle of all time right in front of us, and it's going to take a team effort to solve 🧩🤔 [🎥yt: archivist]

₿en Wehrman

46,363 görüntüleme • 2 ay önce

We ran an SLM fully on an Apple Watch! (the first watch app ever to ship intelligence built into it) My team is obsessed with the question "how small can models go and still be useful?" So Shahir trained a custom model which takes advantage of Apple's Neural Engine hidden inside every Apple Watch. Full autoregressive decoding, close to 30 TPS, ON YOUR WRIST (Series 9)!! 75MB package = 70M param model. Use-Cases? purposefully fine tuning this model to get personal context, health data, run a tiny agent agent for some automation etc. This is exactly what will we be doing next! "Watch out" :P (Also sabesh 📟 exploring alt career options as a hand model)
0:29

We ran an SLM fully on an Apple Watch! (the first watch app ever to ship intelligence built into it) My team is obsessed with the question "how small can models go and still be useful?" So Shahir trained a custom model which takes advantage of Apple's Neural Engine hidden inside every Apple Watch. Full autoregressive decoding, close to 30 TPS, ON YOUR WRIST (Series 9)!! 75MB package = 70M param model. Use-Cases? purposefully fine tuning this model to get personal context, health data, run a tiny agent agent for some automation etc. This is exactly what will we be doing next! "Watch out" :P (Also sabesh 📟 exploring alt career options as a hand model)

Kautuk | Conscious Engines

50,101 görüntüleme • 1 ay önce