Yifan Song

Yifan Song, Mengkun She, Kevin Köser

Underwater image restoration has been a challenging problem for decades since the advent of underwater photography. Most solutions focus on shallow water scenarios, where the scene is uniformly illuminated by the sunlight. However, the vast majority of uncharted underwater terrain is located beyond 200 meters depth where natural light is scarce and artificial illumination is needed. In such cases, light sources co-moving with the camera, dynamically change the scene appearance, which make shallow water restoration methods inadequate. In particular for multi-light source systems (composed of dozens of LEDs nowadays), calibrating each light is time-consuming, error-prone and tedious, and we observe that only the integrated illumination within the viewing volume of the camera is critical, rather than the individual light sources. The key idea of this paper is therefore to exploit the appearance changes of objects or the seafloor, when traversing the viewing frustum of the camera. Through new constraints assuming Lambertian surfaces, corresponding image pixels constrain the light field in front of the camera, and for each voxel a signal factor and a backscatter value are stored in a volumetric grid that can be used for very efficient image restoration of camera-light platforms, which facilitates consistently texturing large 3D models and maps that would otherwise be dominated by lighting and medium artifacts. To validate the effectiveness of our approach, we conducted extensive experiments on simulated and real-world datasets. The results of these experiments demonstrate the robustness of our approach in restoring the true albedo of objects, while mitigating the influence of lighting and medium effects. Furthermore, we demonstrate our approach can be readily extended to other scenarios, including in-air imaging with artificial illumination or other similar cases.

Mengkun She, Yifan Song, David Nakath et al.

Despite impressive results achieved by many on-land visual mapping algorithms in the recent decades, transferring these methods from land to the deep sea remains a challenge due to harsh environmental conditions. Images captured by autonomous underwater vehicles (AUVs), equipped with high-resolution cameras and artificial illumination systems, often suffer from heterogeneous illumination and quality degradation caused by attenuation and scattering, on top of refraction of light rays. These challenges often result in the failure of on-land SLAM approaches when applied underwater or cause SfM approaches to exhibit drifting or omit challenging images. Consequently, this leads to gaps, jumps, or weakly reconstructed areas. In this work, we present a navigation-aided hierarchical reconstruction approach to facilitate the automated robotic 3D reconstruction of hectares of seafloor. Our hierarchical approach combines the advantages of SLAM and global SfM that is much more efficient than incremental SfM, while ensuring the completeness and consistency of the global map. This is achieved through identifying and revisiting problematic or weakly reconstructed areas, avoiding to omit images and making better use of limited dive time. The proposed system has been extensively tested and evaluated during several research cruises, demonstrating its robustness and practicality in real-world conditions.

Dawei Zhu, Liang Wang, Nan Yang et al. · microsoft-research, pku

Embedding models play a pivot role in modern NLP applications such as IR and RAG. While the context limit of LLMs has been pushed beyond 1 million tokens, embedding models are still confined to a narrow context window not exceeding 8k tokens, refrained from application scenarios requiring long inputs such as legal contracts. This paper explores context window extension of existing embedding models, pushing the limit to 32k without requiring additional training. First, we examine the performance of current embedding models for long context retrieval on our newly constructed LongEmbed benchmark. LongEmbed comprises two synthetic tasks and four carefully chosen real-world tasks, featuring documents of varying length and dispersed target information. Benchmarking results underscore huge room for improvement in these models. Based on this, comprehensive experiments show that training-free context window extension strategies like position interpolation can effectively extend the context window of existing embedding models by several folds, regardless of their original context being 512 or beyond 4k. Furthermore, for models employing absolute position encoding (APE), we show the possibility of further fine-tuning to harvest notable performance gains while strictly preserving original behavior for short inputs. For models using rotary position embedding (RoPE), significant enhancements are observed when employing RoPE-specific methods, such as NTK and SelfExtend, indicating RoPE's superiority over APE for context window extension. To facilitate future research, we release E5-Base-4k and E5-RoPE-Base, along with the LongEmbed benchmark.

Jiarui Yang, Songpengcheng Xia, Yifan Song et al.

Human body reconstruction with Millimeter Wave (mmWave) radar point clouds has gained significant interest due to its ability to work in adverse environments and its capacity to mitigate privacy concerns associated with traditional camera-based solutions. Despite pioneering efforts in this field, two challenges persist. Firstly, raw point clouds contain massive noise points, usually caused by the ambient objects and multi-path effects of Radio Frequency (RF) signals. Recent approaches typically rely on prior knowledge or elaborate preprocessing methods, limiting their applicability. Secondly, even after noise removal, the sparse and inconsistent body-related points pose an obstacle to accurate human body reconstruction. To address these challenges, we introduce mmBaT, a novel multi-task deep learning framework that concurrently estimates the human body and predicts body translations in subsequent frames to extract body-related point clouds. Our method is evaluated on two public datasets that are collected with different radar devices and noise levels. A comprehensive comparison against other state-of-the-art methods demonstrates our method has a superior reconstruction performance and generalization ability from noisy raw data, even when compared to methods provided with body-related point clouds.

Mengkun She, Tim Weiß, Yifan Song et al.

Reliable quantification of natural and anthropogenic gas release (e.g.\ CO$_2$, methane) from the seafloor into the water column, and potentially to the atmosphere, is a challenging task. While ship-based echo sounders such as single beam and multibeam systems allow detection of free gas, bubbles, in the water even from a great distance, exact quantification utilizing the hydroacoustic data requires additional parameters such as rise speed and bubble size distribution. Optical methods are complementary in the sense that they can provide high temporal and spatial resolution of single bubbles or bubble streams from close distance. In this contribution we introduce a complete instrument and evaluation method for optical bubble stream characterization targeted at flows of up to 100ml/min and bubbles with a few millimeters radius. The dedicated instrument employs a high-speed deep sea capable stereo camera system that can record terabytes of bubble imagery when deployed at a seep site for later automated analysis. Bubble characteristics can be obtained for short sequences, then relocating the instrument to other locations, or in autonomous mode of definable intervals up to several days, in order to capture bubble flow variations due to e.g. tide dependent pressure changes or reservoir depletion. Beside reporting the steps to make bubble characterization robust and autonomous, we carefully evaluate the reachable accuracy to be in the range of 1-2\% of the bubble radius and propose a novel auto-calibration procedure that, due to the lack of point correspondences, uses only the silhouettes of bubbles. The system has been operated successfully in 1000m water depth at the Cascadia margin offshore Oregon to assess methane fluxes from various seep locations. Besides sample results we also report failure cases and lessons learnt during deployment and method development.

Mengkun She, David Nakath, Yifan Song et al.

Underwater cameras are typically placed behind glass windows to protect them from the water. Spherical glass, a dome port, is well suited for high water pressures at great depth, allows for a large field of view, and avoids refraction if a pinhole camera is positioned exactly at the sphere's center. Adjusting a real lens perfectly to the dome center is a challenging task, both in terms of how to actually guide the centering process (e.g. visual servoing) and how to measure the alignment quality, but also, how to mechanically perform the alignment. Consequently, such systems are prone to being decentered by some offset, leading to challenging refraction patterns at the sphere that invalidate the pinhole camera model. We show that the overall camera system becomes an axial camera, even for thick domes as used for deep sea exploration and provide a non-iterative way to compute the center of refraction without requiring knowledge of exact air, glass or water properties. We also analyze the refractive geometry at the sphere, looking at effects such as forward- vs. backward decentering, iso-refraction curves and obtain a 6th-degree polynomial equation for forward projection of 3D points in thin domes. We then propose a pure underwater calibration procedure to estimate the decentering from multiple images. This estimate can either be used during adjustment to guide the mechanical position of the lens, or can be considered in photogrammetric underwater applications.

Yifan Song, Furkan Elibol, Mengkun She et al.

Illuminating a scene with artificial light is a prerequisite for seeing in dark environments. However, nonuniform and dynamic illumination can deteriorate or even break computer vision approaches, for instance when operating a robot with headlights in the darkness. This paper presents a novel light calibration approach by taking multi-view and -distance images of a reference plane in order to provide pose information of the employed light sources to the computer vision system. By following a physical light propagation approach, under consideration of energy preservation, the estimation of light poses is solved by minimizing of the differences between real and rendered pixel intensities. During the evaluation we show the robustness and consistency of this method by statistically analyzing the light pose estimation results with different setups. Although the results are demonstrated using a rotationally-symmetric non-isotropic light, the method is suited also for non-symmetric lights.