209 research outputs found
Self-Supervised Monocular Depth and Motion Learning in Dynamic Scenes: Semantic Prior to Rescue
No full textWe introduce an end-to-end joint training framework that explicitly models 6-DoF motion of multiple dynamic objects, ego-motion, and depth in a monocular camera setup without geometric supervision. Our technical contributions are three-fold. First, we highlight the fundamental difference between inverse and forward projection while modeling the individual motion of each rigid object, and propose a geometrically correct projection pipeline using a neural forward projection module. Second, we propose two types of residual motion learning frameworks to explicitly disentangle camera and object motions in dynamic driving scenes with different levels of semantic prior knowledge: video instance segmentation as a strong prior, and object detection as a weak prior. Third, we design a unified photometric and geometric consistency loss that holistically imposes self-supervisory signals for every background and object region. Lastly, we present a unsupervised method of 3D motion field regularization for semantically plausible object motion representation. Our proposed elements are validated in a detailed ablation study. Through extensive experiments conducted on the KITTI, Cityscapes, and Waymo open dataset, our framework is shown to outperform the state-of-the-art depth and motion estimation methods. Our code, dataset, and models are publicly available. © 2023 Springer Nature Switzerland AG. Part of Springer Nature.FALSEsciescopu
DPSNet: End-to-end deep plane sweep stereo
No full textMultiview stereo aims to reconstruct scene depth from images acquired by a camera under arbitrary motion. Recent methods address this problem through deep learning, which can utilize semantic cues to deal with challenges such as textureless and reflective regions. In this paper, we present a convolutional neural network called DPSNet (Deep Plane Sweep Network) whose design is inspired by best practices of traditional geometry-based approaches for dense depth reconstruction. Rather than directly estimating depth and/or optical flow correspondence from image pairs as done in many previous deep learning methods, DPSNet takes a plane sweep approach that involves building a cost volume from deep features using the plane sweep algorithm, regularizing the cost volume via a context-aware cost aggregation, and regressing the dense depth map from the cost volume. The cost volume is constructed using a differentiable warping process that allows for end-to-end training of the network. Through the effective incorporation of conventional multiview stereo concepts within a deep learning framework, DPSNet achieves state-of-the-art reconstruction results on a variety of challenging datasets
Learning Monocular Depth in Dynamic Scenes via Instance-Aware Projection Consistency
No full textWe present an end-to-end joint training framework that explicitly models 6-DoF motion of multiple dynamic objects, ego-motion, and depth in a monocular camera setup without supervision. Our technical contributions are three-fold. First, we highlight the fundamental difference between inverse and forward projection while modeling the individual motion of each rigid object, and propose a geometrically correct projection pipeline using a neural forward projection module. Second, we design a unified instance-aware photometric and geometric consistency loss that holistically imposes self-supervisory signals for every background and object region. Lastly, we introduce a general-purpose auto-annotation scheme using any off-the-shelf instance segmentation and optical flow models to produce video instance segmentation maps that will be utilized as input to our training pipeline. These proposed elements are validated in a detailed ablation study. Through extensive experiments conducted on the KITTI and Cityscapes dataset, our framework is shown to outperform the state-of-the-art depth and motion estimation methods. Our code, dataset, and models are publicly available
Resolving the existence of Higgsinos in the LHC inverse problem
No full textThe LHC inverse problem is infamously challenging when neutralinos and charginos are heavy and pure and other superparticles are decoupled. This limit is becoming more relevant to particle physics nowadays. Fortunately, in this limit, Higgsinos produce a distinctive signature if they are the LSPs or NLSPs. The identifying signature is the presence of equal numbers of Z bosons and Higgs bosons in NLSP productions and subsequent decays at hadron colliders. The signature is derived from the Goldstone equivalence theorem by which partial widths into Z and Higgs bosons are inherently related and from the fact that Higgsinos consist of two indistinguishable neutralinos. Thus it is valid in general for many supersymmetry models; exceptions may happen when Higgsino NLSPs decay to weakly coupled LSPs such as axinos or gravitinos. © 2014 The Author(s).Y
Depth Completion with Deep Geometry and Context Guidance
No full textIn this paper, we present an end-to-end convolutional neural network (CNN) for depth completion. Our network consists of a geometry network and a context network. The geometry network, a single encoder-decoder network, learns to optimize a multi-task loss to generate an initial propagated depth map and a surface normal. The complementary outputs allow it to correctly propagate initial sparse depth points in slanted surfaces. The context network extracts a local and a global feature of an image to compute a bilateral weight, which enables it to preserve edges and fine details in the depth maps. At the end, a final output is produced by multiplying the initially propagated depth map with the bilateral weight. In order to validate the effectiveness and the robustness of our network, we performed extensive ablation studies and compared the results against state-of-the-art CNN-based depth completions, where we showed promising results on various scenes
Deep Depth from Uncalibrated Small Motion Clip
No full textWe propose a novel approach to infer a high-quality depth map from a set of images with small viewpoint variations. In general, techniques for depth estimation from small motion consist of camera pose estimation and dense reconstruction. In contrast to prior approaches that recover scene geometry and camera motions using pre-calibrated cameras, we introduce a self-calibrating bundle adjustment method tailored for small motion which enables computation of camera poses without the need for camera calibration. For dense depth reconstruction, we present a convolutional neural network called DPSNet (Deep Plane Sweep Network) whose design is inspired by best practices of traditional geometry-based approaches. Rather than directly estimating depth or optical flow correspondence from image pairs as done in many previous deep learning methods, DPSNet takes a plane sweep approach that involves building a cost volume from deep features using the plane sweep algorithm, regularizing the cost volume, and regressing the depth map from the cost volume. The cost volume is constructed using a differentiable warping process that allows for end-to-end training of the network. Through the effective incorporation of conventional multiview stereo concepts within a deep learning framework, the proposed method achieves state-of-the-art results on a variety of challenging datasets. IEEE1
Very Degenerate Higgsino Dark Matter
No full textWe present a study of the Very Degenerate Higgsino Dark Matter (DM), whose mass splitting between the lightest neutral and charged components is O(1) MeV, much smaller than radiative splitting of 355 MeV. The scenario is realized in the minimal supersymmetric standard model by small gaugino mixings. In contrast to the pure Higgsino DM with the radiative splitting only, various observable signatures with distinct features are induced. First of all, the very small mass splitting makes (a) sizable Sommerfeld enhancement and Ramsauer-Townsend (RT) suppression relevant to ∼1 TeV Higgsino DM, and (b) Sommerfeld-Ramsauer-Townsend effect saturate at lower velocities v/c ≲ 10−3. As a result, annihilation signals can be large enough to be observed from the galactic center and/or dwarf galaxies, while the relative signal sizes can vary depending on the locations of Sommerfeld peaks and RT dips. In addition, at collider experiments, stable chargino signatures can be searched for to probe the model in the future. DM direct detection signals, however, depend on the Wino mass; even no detectable signals can be induced if the Wino is heavier than about 10 TeV. © 2017, The Author(s).Y
Facial Depth and Normal Estimation using Single Dual-Pixel Camera
No full textMany mobile manufacturers recently have adopted Dual-Pixel (DP) sensors in
their flagship models for faster auto-focus and aesthetic image captures.
Despite their advantages, research on their usage for 3D facial understanding
has been limited due to the lack of datasets and algorithmic designs that
exploit parallax in DP images. This is because the baseline of sub-aperture
images is extremely narrow and parallax exists in the defocus blur region. In
this paper, we introduce a DP-oriented Depth/Normal network that reconstructs
the 3D facial geometry. For this purpose, we collect a DP facial data with more
than 135K images for 101 persons captured with our multi-camera structured
light systems. It contains the corresponding ground-truth 3D models including
depth map and surface normal in metric scale. Our dataset allows the proposed
matching network to be generalized for 3D facial depth/normal estimation. The
proposed network consists of two novel modules: Adaptive Sampling Module and
Adaptive Normal Module, which are specialized in handling the defocus blur in
DP images. Finally, the proposed method achieves state-of-the-art performances
over recent DP-based depth/normal estimation methods. We also demonstrate the
applicability of the estimated depth/normal to face spoofing and relighting.Comment: Github page : https://github.com/MinJunKang/DualPixelFace To be
appeared in ECCV 202
Renormalization group-induced phenomena of top pairs from four-quark effective operators
No full textWe study the renormalization group(RG) evolution of four-quark operators that contribute to the top pair production. In particular, we focus on the cases in which certain observables are first induced from the one-loop RG while being absent at tree-level. From the operator mixing pattern, we classify all such RG-induced phenomena and underlying models that can induce them. We then calculate the full one-loop QCD RG evolution as the leading estimator of the effects and address the question of which RG-induced phenomena have largest and observable effects. The answer is related to the color structure of QCD. The studied topics include the RG-induction of top asymmetries, polarizations and polarization mixings as well as issues arising at this order. The RG-induction of top asymmetries is further compared with the generation of asymmetries from QCD and QED at one-loop order. We finally discuss the validity of using the RG as the proxy of one-loop effects on the top pair production. As an aside, we clarify the often-studied relations between top pair observables. © 2014 The Author(s).Y
Asymmetric Aneuploidy in Mesenchymal Stromal Cells Detected by In Situ Karyotyping and Fluorescence In Situ Hybridization: Suggestions for Reference Values for Stem CellsReference Values for Stem Cells
No full textCytogenetic testing is important to ensure patient safety before therapeutic application of mesenchymal stromal cells
(MSCs). However, the standardized methods and criteria for the screening of chromosomal abnormalities of MSCs
have not yet been determined. We investigated the frequency of cytogenetic aberrations in MSCs using G-banding
and fluorescence in situ hybridization (FISH) and suggest reference values for aneuploidy in MSCs. Cytogenetic
analysis was performed on 103 consecutive cultures from 68 MSCs (25 adipose-origin, 20 bone marrow-origin, 18
cord blood-origin, and 5 neural stem cells; 8 from adipose tissue of patients with breast cancer and 60 from healthy
donors). We compared the MSC aneuploidy patterns with those of hematological malignancies and benign hematological
diseases. Interphase FISH showed variable aneuploid clone proportions (1%–20%) in 68 MSCs. The
aneuploidy patterns were asymmetric, and aneuploidy of chromosomes 16, 17, 18, and X occurred most frequently.
Clones with polysomy were significantly more abundant than those with monosomy. The cutoff value of maximum
polysomy rates (upper 95th percentile value) was 13.0%. By G-banding, 5 of the 61 MSCs presented clonal
chromosomal aberrations. Aneuploidy was asymmetric in the malignant hematological diseases, while it was symmetric
in the benign hematological diseases.We suggest an aneuploidy cutoff value of 13%, and FISH for aneuploidy
of chromosomes 16, 17, 18, and X would be informative to evaluate the genetic stability of MSCs. Although it is
unclear whether the aneuploid clones might represent the senescent cell population or transformed cells, more
attention should be focused on the safety of MSCs, and G-banding combined with FISH should be performed. (c) Mary Ann Liebert, Inc.3511sciescopu
- …
