Search for: All records

Video Frame Interpolation aims to recover realistic missing frames between observed frames, generating a highframe- rate video from a low-frame-rate video. However, without additional guidance, the large motion between frames makes this problem ill-posed. Event-based Video Frame Interpolation (EVFI) addresses this challenge by using sparse, high-temporal-resolution event measurements as motion guidance. This guidance allows EVFI methods to significantly outperform frame-only methods. However, to date, EVFI methods have relied on a limited set of paired eventframe training data, severely limiting their performance and generalization capabilities. In this work, we overcome the limited data challenge by adapting pre-trained video diffusion models trained on internet-scale datasets to EVFI. We experimentally validate our approach on real-world EVFI datasets, including a new one that we introduce. Our method outperforms existing methods and generalizes across cameras far better than existing approaches. 

Estimating and disentangling epistemic uncertainty, uncertainty that is reducible with more training data, and aleatoric uncertainty, uncertainty that is inherent to the task at hand, is critically important when applying machine learning to highstakes applications such as medical imaging and weather forecasting. Conditional diffusion models’ breakthrough ability to accurately and efficiently sample from the posterior distribution of a dataset now makes uncertainty estimation conceptually straightforward: One need only train and sample from a large ensemble of diffusion models. Unfortunately, training such an ensemble becomes computationally intractable as the complexity of the model architecture grows. In this work we introduce a new approach to ensembling, hyper-diffusion models (HyperDM), which allows one to accurately estimate both epistemic and aleatoric uncertainty with a single model. Unlike existing single-model uncertainty methods like Monte-Carlo dropout and Bayesian neural networks, HyperDM offers prediction accuracy on par with, and in some cases superior to, multi-model ensembles. Furthermore, our proposed approach scales to modern network architectures such as Attention U-Net and yields more accurate uncertainty estimates compared to existing methods. We validate our method on two distinct real-world tasks: x-ray computed tomography reconstruction and weather temperature forecasting. Source code is publicly available at https://github.com/matthewachan/hyperdm. 

Holographic displays are an upcoming technology for AR and VR applications, with the ability to show 3D content with accurate depth cues, including accommodation and motion parallax. Recent research reveals that only a fraction of holographic pixels are needed to display images with high fidelity, improving energy efficiency in future holographic displays. However, the existing iterative method for computing sparse amplitude and phase layouts does not run in real time; instead, it takes hundreds of milliseconds to render an image into a sparse hologram. In this paper, we present a non-iterative amplitude and phase computation for sparse Fourier holograms that uses Perlin noise in the image–plane phase. We conduct simulated and optical experiments. Compared to the Gaussian-weighted Gerchberg–Saxton method, our method achieves a run time improvement of over 600 times while producing a nearly equal PSNR and SSIM quality. The real-time performance of our method enables the presentation of dynamic content crucial to AR and VR applications, such as video streaming and interactive visualization, on holographic displays. 

The rapidly increasing capabilities of autonomous mobile robots promise to make them ubiquitous in the coming decade. These robots will continue to enhance efficiency and safety in novel applications such as disaster management, environmental monitoring, bridge inspection, and agricultural inspection. To operate autonomously without constant human intervention, even in remote or hazardous areas, robots must sense, process, and interpret environmental data using only onboard sensing and computation. This capability is made possible by advancements in perception algorithms, allowing these robots to rely primarily on their perception capabilities for navigation tasks. However, tiny robot autonomy is hindered mainly by sensors, memory, and computing due to size, area, weight, and power constraints. The bottleneck in these robots lies in the real-time perception in resource-constrained robots. To enable autonomy in robots of sizes that are less than 100 mm in body length, we draw inspiration from tiny organisms such as insects and hummingbirds, known for their sophisticated perception, navigation, and survival abilities despite their minimal sensor and neural system. This work aims to provide insights into designing a compact and efficient minimal perception framework for tiny autonomous robots from higher cognitive to lower sensor levels. 

Differentiable 3D-Gaussian splatting (GS) is emerging as a prominent technique in computer vision and graphics for reconstructing 3D scenes. GS represents a scene as a set of 3D Gaussians with varying opacities and employs a computationally efficient splatting operation along with analytical derivatives to compute the 3D Gaussian parameters given scene images captured from various viewpoints. Unfortunately, capturing surround view (360° viewpoint) images is impossible or impractical in many real-world imaging scenarios, including underwater imaging, rooms inside a building, and autonomous navigation. In these restricted baseline imaging scenarios, the GS algorithm suffers from a well-known ‘missing cone’ problem, which results in poor reconstruction along the depth axis. In this paper, we demonstrate that using transient data (from sonars) allows us to address the missing cone problem by sampling high-frequency data along the depth axis. We extend the Gaussian splatting algorithms for two commonly used sonars and propose fusion algorithms that simultaneously utilize RGB camera data and sonar data. Through simulations, emulations, and hardware experiments across various imaging scenarios, we show that the proposed fusion algorithms lead to significantly better novel view synthesis (5 dB improvement in PSNR) and 3D geometry reconstruction (60% lower Chamfer distance).