Papers

Imagine you are learning a second language. You spend months mastering French. Then, you switch gears to learn Spanish. A few months later, you try to speak French again, but you find yourself inserting Spanish words or, worse, you’ve forgotten the French grammar entirely. This phenomenon, known in psychology as catastrophic forgetting, is a massive headache for Artificial Intelligence, specifically for Multimodal Large Language Models (MLLMs). These models, like GPT-4V or Gemini, are incredibly powerful at understanding images and answering questions about them. However, the world changes constantly. We want these models to learn new types of data and tasks continuously without having to retrain them from scratch (which costs millions of dollars) and—crucially—without them forgetting what they learned previously. ...

Depth estimation—the ability to look at a 2D image and understand the 3D geometry within it—is a cornerstone of computer vision. It is the prerequisite for autonomous driving, robot navigation, mixed reality, and content generation. However, building an “ideal” depth estimation model has historically been a game of trade-offs. You usually have to pick two of the following three: Meticulous Detail: Can the model see the fine edges of leaves or the texture of a distant building? Temporal Consistency: If applied to a video, does the depth map flicker, or is it stable over time? Efficiency: Can it run in real-time on a robot, or does it take several seconds per frame? Recent foundation models like Marigold or Depth Anything have pushed the boundaries of detail, but often at the cost of speed or video stability. In this post, we will explore CH3Depth, a new research paper that proposes a unified framework to solve this trilemma. Using a technique called Flow Matching combined with a novel sampling strategy, CH3Depth achieves state-of-the-art results in both image and video depth estimation while being significantly faster than its predecessors. ...

Introduction Imagine you are shopping online for a shirt. You find a photo of a shirt that has the perfect cut and fabric, but it’s blue, and you really wanted it in grey. In a standard text search, describing exactly what you want is difficult (“shirt like this but grey”). This scenario is where Composed Image Retrieval (CIR) shines. CIR allows users to search using a combination of a reference image (the blue shirt) and a text instruction (“change blue to grey”). Ideally, the system understands that it should keep the shape and fabric from the image but swap the color based on the text. ...

Introduction In the rapidly evolving world of Artificial Intelligence, Large Multimodal Models (LMMs) have emerged as the new titans. Models like LLaVA and GPT-4V can see, read, and reason, bridging the gap between visual and textual data. However, this capability comes at a steep price: computational resources. To put this into perspective, running a 70-billion parameter model like LLaVA-Onevision at standard 16-bit precision requires roughly 140GB of GPU memory. This effectively walls off these powerful models from consumer hardware and efficient edge deployment. To solve this, researchers turn to model compression, specifically quantization—reducing the precision of the model’s weights (e.g., from 16-bit floating point to 4-bit or 2-bit integers). ...

Introduction Imagine you are trying to furnish a virtual apartment. You place a stylish L-shaped sofa in the corner and a coffee table in the nook of the “L”. To you, this is a perfect, cozy arrangement. But to a computer using traditional 3D understanding, you might have just caused a catastrophe. Why? Because many computer vision models view objects not as complex shapes, but as simple bounding boxes. If the bounding box of the table touches the bounding box of the sofa—even if the actual objects don’t touch—the computer screams “Collision!” Conversely, models might generate scenes where objects physically overlap in reality because the coarse boxes allowed it. ...

Unlocking Mobile Vision: How CARE Transformers Balance Speed and Accuracy In the rapidly evolving world of Computer Vision, the Vision Transformer (ViT) has been a revolutionary force. By adapting the self-attention mechanisms originally designed for Natural Language Processing (NLP), ViTs have achieved state-of-the-art results in image classification, object detection, and segmentation. However, there is a catch. The very mechanism that makes Transformers so powerful—Self-Attention—is computationally expensive. Specifically, it has “quadratic complexity.” As the resolution of an image increases, the computational cost explodes. This makes standard Transformers notoriously difficult to deploy on resource-constrained devices like mobile phones, where battery life and latency are critical. ...

Breaking the Parts Barrier: How CAP-Net Masters Articulated Object Perception Imagine you are a robot tasked with a seemingly simple household chore: opening a laptop. To a human, this is trivial. You identify the lid, find the edge, and lift. But to a robot, this is a geometric nightmare. The laptop is not a solid brick; it is an articulated object—a structure composed of rigid parts connected by joints. The lid moves relative to the base, changing the object’s overall shape. ...

Introduction: The Gap Between AI Art and Engineering In the last few years, generative AI has transformed how we visualize ideas. Tools like Midjourney or Stable Diffusion can conjure photorealistic scenes from a text prompt, and recent breakthroughs in 3D generation—like DreamFusion or Wonder3D—can turn a single 2D image into a rotating 3D asset. However, if you are an engineer, a product designer, or a game developer, you likely face a frustrating reality: generated 3D meshes are often useless for manufacturing. ...

How to Train CLIP with Infinite Batch Sizes: Breaking the Memory Barrier In the world of modern AI, specifically in Representation Learning, there is a recurring theme: bigger is usually better. This is particularly true for contrastive learning models like CLIP (Contrastive Language-Image Pre-training). The secret sauce behind these models isn’t just the architecture; it’s the data, and more importantly, how much data the model sees at once. Research has consistently shown that larger batch sizes lead to better performance. A larger batch provides a more diverse set of “negative” samples (images that don’t match the text), forcing the model to learn much sharper, more discriminative features. ...

Introduction We have all seen them: AI-generated portraits that look almost right, but something is off. Perhaps the skin texture is too plastic, the eyes lack a certain spark, or the anatomy twists in ways human bones simply shouldn’t. Despite the massive leaps in diffusion models like Stable Diffusion, generating truly photorealistic humans remains one of the hardest challenges in computer vision. The core issue often lies in how these models are fine-tuned. Typically, researchers use methods like Reinforcement Learning from Human Feedback (RLHF) or Direct Preference Optimization (DPO). These methods train the model by showing it two generated images—one “good” and one “bad”—and telling it to prefer the good one. But there is a ceiling to this approach: if the model’s “good” image is still artificial and flawed, the model is only learning to be the “best of a bad bunch.” It isn’t learning what real looks like. ...

In the world of computer vision and robotics, 3D reconstruction is the holy grail. Whether it’s a robot navigating a warehouse, a VR headset mapping your living room, or a Mars rover scanning a dune, the ability to turn the real world into a digital 3D model is critical. For years, the gold standard for handheld scanning (like what you might find on a high-end smartphone) has been a combination of an RGB camera and a sparse LiDAR sensor. This setup works reasonably well in perfect conditions. But the real world isn’t perfect. We encounter dark rooms, white textureless walls, and black objects that absorb light. In these “challenging” scenarios, traditional RGB-based reconstruction fails because it can’t “see” features, and sparse LiDAR fails because it doesn’t capture enough data points to fill in the gaps. ...

The world of 3D graphics—the backbone of modern video games, blockbuster movies, and architectural visualization—is notoriously complex. Creating a photorealistic scene isn’t just about artistic vision; it requires technical mastery of sophisticated software like Blender, Maya, or Unreal Engine. An artist doesn’t just “draw” a 3D chair; they manipulate geometry nodes, adjust material shaders, tweak lighting coordinates, and wrangle with physics simulations. Because this process is so time-consuming and specialized, researchers have been racing to automate it using Artificial Intelligence. We’ve seen the rise of Vision-Language Models (VLMs) that can look at an image and understand what’s in it. The dream is simple: tell an AI, “Make the lights dimmer and turn that wooden table into glass,” and have it execute the task instantly. ...

Balancing Act: How Multi-Objective Optimization Boosts Learned Image Compression In the world of digital media, we are constantly fighting a tug-of-war. On one side, we want high-quality images that look crisp and true to life (Low Distortion). On the other, we want files that are small enough to stream, store, and share instantly (Low Rate). This trade-off is the heart of image compression. Traditional codecs like JPEG or HEVC solve this with hand-tuned engineering. But recently, Learned Image Compression (LIC)—using deep neural networks to compress images—has started to outperform these traditional methods. LIC models learn from data how to best represent an image. ...

Introduction Imagine trying to draw a precise blueprint of a house, but all you have is a grainy, satellite-like scan taken from a plane flying overhead. Some parts of the roof are missing, trees are blocking the walls, and the data is just a collection of scattered dots. This is the reality of reconstructing 3D building models from airborne LiDAR (Light Detection and Ranging) point clouds. Building reconstruction is a cornerstone technology for smart cities, autonomous driving, and Virtual Reality/Augmented Reality (VR/AR). While we have become good at capturing data, turning that raw, noisy data into clean, lightweight “wireframes” (skeletal representations of geometry) remains a massive challenge. ...

Introduction Imagine driving down a highway at night in pouring rain. Your eyes strain to distinguish between a parked car on the shoulder and a shadow, or between a distant streetlight and an oncoming vehicle. Now, imagine you are a computer algorithm trying to do the same thing. In autonomous driving, the vehicle’s “brain” typically relies on a Bird’s-Eye-View (BEV) representation. This is a top-down, grid-like map of the surroundings generated from onboard cameras and LiDAR sensors. This map is the foundation for everything the car does next: detecting objects, predicting movement, and planning a path. ...

Introduction Imagine you are standing in the middle of a room, holding a camera, and taking a \(360^{\circ}\) panoramic photo. Now, walk into the next room and take another one. Can you reconstruct the entire floor plan of the house—accurate to the centimeter—just from those two photos? This is the problem of Wide-Baseline Floor Plan Reconstruction, and it is notoriously difficult. Unlike a video feed where camera frames are millimeters apart, wide-baseline images are taken far apart (often in different rooms). The visual overlap is small, and traditional computer vision algorithms struggle to stitch these “islands” of data together into a coherent map. ...

For the last few years, the recipe for building a Vision-Language Model (VLM) has been relatively static. If you wanted a model that understood how images and text relate—like OpenAI’s CLIP—you needed to collect a massive dataset of hundreds of millions of image-text pairs and train two neural networks (one for vision, one for text) from scratch. This process is computationally expensive, data-hungry, and often results in models that are “jacks of all trades, masters of none.” The vision encoder might be decent, and the text encoder might be passable, but neither is state-of-the-art compared to models dedicated to a single modality. ...

From Chaos to Code: Transforming Sparse Point Clouds into Structured 3D Buildings with ArcPro Imagine flying a drone over a city to map it. The drone captures thousands of images, and through photogrammetry, you generate a 3D representation of the scene. What you get back, however, is rarely a pristine, CAD-ready model. Instead, you get a “point cloud”—a chaotic swarm of millions of floating dots. If the scan is high-quality, the dots are dense, and you can see the surfaces clearly. But in the real world, data is often messy. Aerial scans can be sparse (containing very few points), noisy (points are in the wrong place), or incomplete (entire walls might be missing due to occlusion). ...

Introduction In the world of Computer Vision, things are surprisingly orderly. Whether you are training a model on ImageNet or your own collection of vacation photos, the data usually looks the same: standard RGB images, captured by standard cameras, often resized to a standard resolution (like \(224 \times 224\)). This uniformity has allowed models like ResNet and Vision Transformers (ViTs) to become powerful, general-purpose engines. But if you look at the planet from space, that order collapses into chaos. ...

Introduction In the world of medical diagnostics, there is rarely a single, indisputable truth. When three different radiologists look at a CT scan of a lung nodule or an MRI of a tumor, they will likely draw three slightly different boundaries around the lesion. This isn’t an error; it is the inherent ambiguity of medical imaging caused by blurred edges, low contrast, and complex anatomy. However, traditional Deep Learning models treat segmentation as a deterministic task. They are trained to output a single “correct” mask. This creates a disconnect between AI outputs and clinical reality. Furthermore, training these models requires massive datasets with pixel-perfect annotations, which are incredibly expensive and time-consuming to obtain. ...