CVPR 2025

Imagine you show a child a red apple. They learn what an “apple” is. Later, you show them a green apple, or perhaps a plastic toy apple painted blue. The child immediately recognizes it as an apple because they understand its shape and structure, not just its color or texture. Now, try the same experiment with a standard computer vision model. If trained only on red apples, many models will fail spectacularly when presented with a blue one. Why? Because deep neural networks are notoriously lazy: they often cheat by memorizing textures (like the specific shiny red skin) rather than learning the underlying geometry of the object. ...

Introduction In the rapidly evolving world of Computer Vision, there is a prevailing tendency to solve complex problems by adding complexity to architectures. When the Vision Transformer (ViT) burst onto the scene, it revolutionized image classification. However, when researchers tried to apply it to more granular tasks like image segmentation—where the goal is to classify every pixel—the consensus was that the “plain” ViT wasn’t enough. To bridge the gap, the field developed a standard recipe: take a ViT, attach a heavy “adapter” to extract multi-scale features (mimicking Convolutional Neural Networks), add a “pixel decoder” to fuse these features, and finally, cap it off with a complex “Transformer decoder” to generate masks. State-of-the-art models like Mask2Former follow this pattern, becoming powerful but architecturally convoluted and computationally heavy. ...

Introduction In the rapidly evolving landscape of Artificial Intelligence, Large Vision-Language Models (LVLMs) like LLaVA, GPT-4V, and DeepSeek-VL have become the superstars of multimodal understanding. These models possess an uncanny ability to describe complex scenes, answer questions about images, and even engage in reasoning tasks that were previously thought impossible. However, there has been a persistent gap in their capabilities. While an LVLM can eloquently describe a “red car parked next to a fire hydrant,” asking it to pinpoint the exact pixel coordinates of that car often results in failure or requires significant modifications to the model. This task—identifying the specific region in an image that corresponds to a text description—is known as Visual Grounding. ...

Introduction How do you understand the three-dimensional structure of the world? You don’t walk around with a ruler and a protractor, measuring the precise coordinates of every object you see. You don’t rely on “gold-standard” 3D mesh data implanted in your brain. Instead, you observe. You move your head, walk around a statue, or drive down a street. Your brain stitches these continuous 2D observations into a coherent 3D model. ...

We live in an era where Multimodal Large Language Models (MLLMs) like GPT-4o and Gemini can describe a photo of a cat on a couch with poetic detail. They can explain memes, read charts, and even help fix your coding errors based on a screenshot. But what happens when you ask these same models to look at the world from 20,000 feet up? Imagine asking an AI to analyze a satellite image of a sprawling port city. You don’t just want to know “is this a city?” You want to know: How many ships are docked? Is that traffic jam near the bridge caused by an accident? What is the safest route for a truck to get from the warehouse to the pier? ...

The dream of “Harry Potter”-style moving photographs has been a driving force in computer vision for decades. We want to take a single static photo of a person and animate it using a driving video—making the subject dance, speak, or walk while preserving their identity. While recent advances in diffusion models have made this possible, there is a lingering “uncanny valley” effect in current state-of-the-art methods. You might see a person dancing perfectly, but their hair behaves like a solid helmet, their dress moves like rigid cardboard, and the background remains frozen in time. The person moves, but the dynamics—the physics of wind, gravity, and momentum—are missing. ...

The recent explosion in generative AI has given us models capable of dreaming up distinctive images and surreal videos from simple text prompts. We have seen tremendous progress with diffusion models, which have evolved from generating static portraits to synthesizing dynamic short films. However, if you look closely at AI-generated videos, you will often notice a subtle, nagging problem: the world doesn’t always stay “solid.” Objects might slightly warp as the camera moves; the geometry of a room might shift impossibly; or the background might hallucinate new details that contradict previous frames. This happens because most video diffusion models are learning pixel consistency over time, but they don’t inherently understand the 3D structure of the world they are rendering. They are excellent 2D artists, but poor 3D architects. ...

Imagine you are playing a video game or designing a virtual environment. You snap a photo of a street corner, and you want that photo to instantly expand into a fully navigable, infinite 3D world. You want to walk down that street, turn the corner, and see new buildings, parks, and skies generated in real-time, exactly as you imagine them. For years, this has been the “holy grail” of computer vision and graphics. While we have seen massive leaps in generative AI (like Midjourney for 2D images) and 3D reconstruction (like NeRFs and Gaussian Splatting), combining them into a fast, interactive experience has remained elusive. Current methods are typically “offline”—meaning you provide an image, wait 30 minutes to an hour for a server to process it, and get back a static 3D scene. ...

Introduction Imagine you are trying to learn how to repair a bicycle wheel or perfect a basketball jump shot. You find a video tutorial online, but it’s not just a standard video—it’s a multi-view experience recorded by five different cameras. One camera is strapped to the instructor’s head (egocentric), while four others are placed on tripods around the room (exocentric). This rich data is fantastic for capturing every detail, but it presents a massive cognitive load. As a viewer, you cannot watch five screens simultaneously. You need a director—someone (or something) to switch to the “best” view at every moment. When the mechanic is tightening a spoke, you want the close-up of their hands. When the basketball player is driving to the hoop, you want the wide angle showing the court. ...

Introduction In the world of computer vision, data is the new oil, but refining that oil—specifically, annotating images—is incredibly expensive. This is particularly true for Instance Segmentation, the task of identifying and outlining every distinct object in an image at the pixel level. Unlike simple bounding boxes or image tags, creating a precise mask for every pedestrian, car, or cup in a dataset requires significant human effort and time. To solve this, researchers have turned to Weakly Supervised Instance Segmentation (WSIS). The goal of WSIS is to train models that can predict pixel-perfect masks while only using “cheap” labels during training. These cheap labels typically fall into three categories: ...

Introduction In the fast-moving world of neural rendering, we are often forced to choose between two paths: physical accuracy or rendering speed. On one side, we have ray-tracing methods like NeRF (Neural Radiance Fields). They meticulously simulate light passing through a volume, integrating density along rays. They are physically grounded and produce stunningly realistic images, but they can be painfully slow to train and render. On the other side, we have the recent superstar: 3D Gaussian Splatting (3DGS). It is blazingly fast because it treats the scene as a collection of 3D ellipsoids and “splats” them onto the screen using rasterization. However, this speed comes with a hidden cost. To make the math work for real-time rasterization, 3DGS makes several mathematical approximations. It essentially flattens 3D shapes into 2D stains on your screen, breaking the laws of volume rendering physics. ...

Introduction In the world of cell biology, seeing is understanding. 3D fluorescence confocal (FC) microscopy has become an indispensable tool for scientists, allowing them to peer inside living organisms and visualize the complex, volumetric dance of life at a cellular level. From studying how embryos develop to understanding neural connections, the ability to capture 3D data is revolutionary. However, this technology comes with a frustrating trade-off. To keep cells alive during long-term observation, scientists must keep the laser power low. Low power means less signal, which inevitably leads to noisy, grainy images. Furthermore, the physics of microscopes creates a specific problem known as anisotropic resolution. While the image might look sharp in the lateral plane (XY), the resolution along the depth axis (Z) is often terrible—sometimes 4.5 times worse. This results in 3D volumes that look like stacks of pancakes rather than continuous, solid objects. ...

Introduction Imagine you are using a generative AI tool to edit a photo of a young person. You adjust the “Age” slider to make them look older. The model successfully adds wrinkles and greys the hair, but strangely, it also puts a pair of glasses on the person’s face. You didn’t ask for glasses. You try again, and the same thing happens. Why does this happen? The answer lies in spurious correlations. In many training datasets (like CelebA), older people are statistically more likely to wear glasses than younger people. A standard deep learning model doesn’t understand biology or optics; it simply memorizes patterns. It learns that “Old” and “Glasses” often go together, so when you ask for one, it gives you both. ...

Imagine taking two casual photos of a room—perhaps one of the desk and one of the bookshelf—and instantly generating a fully navigable 3D video of the entire space. No expensive scanning equipment, no hours of processing time, and no “hallucinated” geometry where walls warp into furniture. This is the “Holy Grail” of computer vision: Sparse-view 3D reconstruction. While recent advancements in AI video generators (like Sora) are impressive, they struggle with this specific task. They often lack 3D consistency—meaning as the camera moves, the shape of the room might subtly morph. Furthermore, they are slow, requiring dozens of denoising steps to produce a single second of video. ...

Introduction In the race toward Artificial General Intelligence (AGI), Multimodal Large Language Models (MLLMs) have taken center stage. We have seen models like GPT-4V and Gemini demonstrate incredible proficiency in understanding static images—describing complex scenes, reading handwriting, and even explaining memes. However, the real world is not a series of frozen snapshots; it is a dynamic, continuous flow of visual, auditory, and textual information. To truly approximate human-level perception, AI must master video analysis. But here lies a significant gap: while MLLM development has surged, the benchmarks used to test them have lagged behind. Most existing video benchmarks focus on short clips (often just a few seconds long) or lack the diverse data modalities (like subtitles and audio) that make video such a rich medium. ...

Introduction In the world of computer vision, estimating depth from a single image—determining how far away every pixel is—has seen revolutionary progress. Models like Depth Anything V2 can look at a flat photograph and intuitively understand the 3D geometry of the scene with remarkable accuracy. However, a massive gap remains between understanding a static image and understanding a video. If you simply run a standard image depth model on a video, frame by frame, you encounter a phenomenon known as “flickering.” Because the model processes each frame in isolation, slight changes in lighting or camera angle cause the predicted depth to jump erratically. The result is a jittery, inconsistent mess that is unusable for robotics, augmented reality, or video editing. ...

Introduction In the rapidly evolving world of Artificial Intelligence, we have reached a fascinating recursive milestone: we are increasingly relying on AI models to evaluate other AI models. As Large Vision-Language Models (LVLMs) like GPT-4o and Claude 3.5 Sonnet become more capable, human evaluation becomes prohibitively expensive and slow. To solve this, researchers use “Generative Reward Models” (GenRMs)—essentially using a powerful LVLM as a judge to rank responses, provide feedback, and guide the training of newer models through Reinforcement Learning from Human Feedback (RLHF). ...

Introduction: The Paradox of the Generalist In the rapid evolution of Artificial Intelligence, we have seen the rise of massive “Generalist” Vision-Language Models (VLMs) like GPT-4o and Gemini. These models are incredibly impressive—they can write poetry, analyze charts, and even joke about a photograph. However, when it comes to high-stakes fields like healthcare, being a “jack of all trades” often means being a master of none. A generalist VLM might look at a chest X-ray and correctly identify the lungs, but fail to notice a subtle fracture or a developing tumor that a trained radiologist would spot instantly. Why? Because these models rely on memorized internet knowledge rather than deep, domain-specific visual expertise. They are prone to hallucinations, confidently stating medical facts that are simply wrong. ...

When we watch a movie, we don’t just see a sequence of images; we see a story told through a specific language. A “low-angle shot” makes a character look powerful. A “smash cut” creates sudden shock. A “match cut” draws a thematic connection between two different times or places. As humans, we intuitively understand this visual grammar. However, for Video Large Language Models (Vid-LLMs)—the AI systems designed to understand video content—this “grammar” of film has largely been a foreign language. While modern AI has become exceptionally good at identifying what is happening in a video (e.g., “a man is running”), it has historically struggled to understand how the video is constructed (e.g., “the camera is tracking the man with a handheld shake to imply urgency”). ...

Introduction In the rapidly evolving landscape of Artificial Intelligence, generative models—from GANs to Diffusion models—have become incredibly adept at creating realistic images. When researchers release a new model, they typically attach a scorecard: a Fréchet Inception Distance (FID) or an Inception Score (IS). These metrics provide a single number indicating how “good” the generated images are compared to a reference dataset. But a single number hides a multitude of sins. Two models might have the same FID score but fail in completely different ways. One might refuse to generate cats; the other might generate only cats. One might memorize the training data; the other might hallucinate wildly. Quantitative scores cannot answer the qualitative question: “What specifically is this model generating differently compared to the reference data?” ...