New submissions for Thu, 7 Apr 22

[zhuhu00]

Keyword: SLAM

There is no result

Keyword: Visual inertial

There is no result

Keyword: livox

There is no result

Keyword: loam

There is no result

Keyword: Visual inertial odometry

There is no result

Keyword: lidar

There is no result

Keyword: loop detection

There is no result

Keyword: autonomous driving

Rolling Colors: Adversarial Laser Exploits against Traffic Light Recognition

• Authors: Chen Yan, Zhijian Xu, Zhanyuan Yin, Xiaoyu Ji, Wenyuan Xu
• Subjects: Computer Vision and Pattern Recognition (cs.CV); Cryptography and Security (cs.CR)
• Arxiv link: https://arxiv.org/abs/2204.02675
• Pdf link: https://arxiv.org/pdf/2204.02675
• Abstract Traffic light recognition is essential for fully autonomous driving in urban areas. In this paper, we investigate the feasibility of fooling traffic light recognition mechanisms by shedding laser interference on the camera. By exploiting the rolling shutter of CMOS sensors, we manage to inject a color stripe overlapped on the traffic light in the image, which can cause a red light to be recognized as a green light or vice versa. To increase the success rate, we design an optimization method to search for effective laser parameters based on empirical models of laser interference. Our evaluation in emulated and real-world setups on 2 state-of-the-art recognition systems and 5 cameras reports a maximum success rate of 30% and 86.25% for Red-to-Green and Green-to-Red attacks. We observe that the attack is effective in continuous frames from more than 40 meters away against a moving vehicle, which may cause end-to-end impacts on self-driving such as running a red light or emergency stop. To mitigate the threat, we propose redesigning the rolling shutter mechanism.

  Keyword: mapping

  Pareto-optimal clustering with the primal deterministic information bottleneck

• Authors: Andrew K. Tan, Max Tegmark, Isaac L. Chuang
• Subjects: Machine Learning (cs.LG); Information Theory (cs.IT); Machine Learning (stat.ML)
• Arxiv link: https://arxiv.org/abs/2204.02489
• Pdf link: https://arxiv.org/pdf/2204.02489
• Abstract At the heart of both lossy compression and clustering is a trade-off between the fidelity and size of the learned representation. Our goal is to map out and study the Pareto frontier that quantifies this trade-off. We focus on the Deterministic Information Bottleneck (DIB) formulation of lossy compression, which can be interpreted as a clustering problem. To this end, we introduce the {\it primal} DIB problem, which we show results in a much richer frontier than its previously studied dual counterpart. We present an algorithm for mapping out the Pareto frontier of the primal DIB trade-off that is also applicable to most other two-objective clustering problems. We study general properties of the Pareto frontier, and give both analytic and numerical evidence for logarithmic sparsity of the frontier in general. We provide evidence that our algorithm has polynomial scaling despite the super-exponential search space; and additionally propose a modification to the algorithm that can be used where sampling noise is expected to be significant. Finally, we use our algorithm to map the DIB frontier of three different tasks: compressing the English alphabet, extracting informative color classes from natural images, and compressing a group theory inspired dataset, revealing interesting features of frontier, and demonstrating how the structure of the frontier can be used for model selection with a focus on points previously hidden by the cloak of the convex hull.

  Cloning Outfits from Real-World Images to 3D Characters for Generalizable Person Re-Identification

• Authors: Yanan Wang, Xuezhi Liang, Shengcai Liao
• Subjects: Computer Vision and Pattern Recognition (cs.CV)
• Arxiv link: https://arxiv.org/abs/2204.02611
• Pdf link: https://arxiv.org/pdf/2204.02611
• Abstract Recently, large-scale synthetic datasets are shown to be very useful for generalizable person re-identification. However, synthesized persons in existing datasets are mostly cartoon-like and in random dress collocation, which limits their performance. To address this, in this work, an automatic approach is proposed to directly clone the whole outfits from real-world person images to virtual 3D characters, such that any virtual person thus created will appear very similar to its real-world counterpart. Specifically, based on UV texture mapping, two cloning methods are designed, namely registered clothes mapping and homogeneous cloth expansion. Given clothes keypoints detected on person images and labeled on regular UV maps with clear clothes structures, registered mapping applies perspective homography to warp real-world clothes to the counterparts on the UV map. As for invisible clothes parts and irregular UV maps, homogeneous expansion segments a homogeneous area on clothes as a realistic cloth pattern or cell, and expand the cell to fill the UV map. Furthermore, a similarity-diversity expansion strategy is proposed, by clustering person images, sampling images per cluster, and cloning outfits for 3D character generation. This way, virtual persons can be scaled up densely in visual similarity to challenge model learning, and diversely in population to enrich sample distribution. Finally, by rendering the cloned characters in Unity3D scenes, a more realistic virtual dataset called ClonedPerson is created, with 5,621 identities and 887,766 images. Experimental results show that the model trained on ClonedPerson has a better generalization performance, superior to that trained on other popular real-world and synthetic person re-identification datasets. The ClonedPerson project is available at https://github.com/Yanan-Wang-cs/ClonedPerson.

  There Are a Thousand Hamlets in a Thousand People's Eyes: Enhancing Knowledge-grounded Dialogue with Personal Memory

• Authors: Tingchen Fu, Xueliang Zhao, Chongyang Tao, Ji-Rong Wen, Rui Yan
• Subjects: Computation and Language (cs.CL)
• Arxiv link: https://arxiv.org/abs/2204.02624
• Pdf link: https://arxiv.org/pdf/2204.02624
• Abstract Knowledge-grounded conversation (KGC) shows great potential in building an engaging and knowledgeable chatbot, and knowledge selection is a key ingredient in it. However, previous methods for knowledge selection only concentrate on the relevance between knowledge and dialogue context, ignoring the fact that age, hobby, education and life experience of an interlocutor have a major effect on his or her personal preference over external knowledge. Without taking the personalization issue into account, it is difficult to select the proper knowledge and generate persona-consistent responses. In this work, we introduce personal memory into knowledge selection in KGC to address the personalization issue. We propose a variational method to model the underlying relationship between one's personal memory and his or her selection of knowledge, and devise a learning scheme in which the forward mapping from personal memory to knowledge and its inverse mapping is included in a closed loop so that they could teach each other. Experiment results show that our method outperforms existing KGC methods significantly on both automatic evaluation and human evaluation.

  Exploration with Global Consistency Using Real-Time Re-integration and Active Loop Closure

• Authors: Yichen Zhang, Boyu Zhou, Luqi Wang, Shaojie Shen
• Subjects: Robotics (cs.RO)
• Arxiv link: https://arxiv.org/abs/2204.02693
• Pdf link: https://arxiv.org/pdf/2204.02693
• Abstract Despite recent progress of robotic exploration, most methods assume that drift-free localization is available, which is problematic in reality and causes severe distortion of the reconstructed map. In this work, we present a systematic exploration mapping and planning framework that deals with drifted localization, allowing efficient and globally consistent reconstruction. A real-time re-integration-based mapping approach along with a frame pruning mechanism is proposed, which rectifies map distortion effectively when drifted localization is corrected upon detecting loop-closure. Besides, an exploration planning method considering historical viewpoints is presented to enable active loop closing, which promotes a higher opportunity to correct localization errors and further improves the mapping quality. We evaluate both the mapping and planning methods as well as the entire system comprehensively in simulation and real-world experiments, showing their effectiveness in practice. The implementation of the proposed method will be made open-source for the benefit of the robotics community.

  DiffSearch: A Scalable and Precise Search Engine for Code Changes

• Authors: Luca Di Grazia, Paul Bredl, Michael Pradel
• Subjects: Software Engineering (cs.SE)
• Arxiv link: https://arxiv.org/abs/2204.02787
• Pdf link: https://arxiv.org/pdf/2204.02787
• Abstract The source code of successful projects is evolving all the time, resulting in hundreds of thousands of code changes stored in source code repositories. This wealth of data can be useful, e.g., to find changes similar to a planned code change or examples of recurring code improvements. This paper presents DiffSearch, a search engine that, given a query that describes a code change, returns a set of changes that match the query. The approach is enabled by three key contributions. First, we present a query language that extends the underlying programming language with wildcards and placeholders, providing an intuitive way of formulating queries that is easy to adapt to different programming languages. Second, to ensure scalability, the approach indexes code changes in a one-time preprocessing step, mapping them into a feature space, and then performs an efficient search in the feature space for each query. Third, to guarantee precision, i.e., that any returned code change indeed matches the given query, we present a tree-based matching algorithm that checks whether a query can be expanded to a concrete code change. We present implementations for Java, JavaScript, and Python, and show that the approach responds within seconds to queries across one million code changes, has a recall of 80.7% for Java, 89.6% for Python, and 90.4% for JavaScript, enables users to find relevant code changes more effectively than a regular expression-based search, and is helpful for gathering a large-scale dataset of real-world bug fixes.

  SPIDER-WEB enables stable, repairable, and encryptible algorithms under arbitrary local biochemical constraints in DNA-based storage

• Authors: Haoling Zhang, Zhaojun Lan, Wenwei Zhang, Xun Xu, Zhi Ping, Yiwei Zhang, Yue Shen.
• Subjects: Emerging Technologies (cs.ET); Information Theory (cs.IT); Combinatorics (math.CO); Genomics (q-bio.GN)
• Arxiv link: https://arxiv.org/abs/2204.02855
• Pdf link: https://arxiv.org/pdf/2204.02855
• Abstract DNA has become an attractive medium for storing digital information gradually. Despite the biochemical progress on DNA synthesis and sequencing, novel coding algorithms need to be constructed under the specific constraints in DNA-based storage. In recent years, a growing number of functional operations and storage carriers were introduced, bringing in various biochemical constraints including but not confined to long single-nucleotide repeats and abnormal GC content. Given several local biochemical constraints and their combinations, existing coding algorithms are not applicable or unstable. In this paper, we design a graph-based architecture, named SPIDER-WEB, to generate corresponding graph-based algorithms under arbitrary local biochemical constraints. These generated coding algorithms could be used to encode arbitrary digital data as DNA sequences directly or served as a benchmark for the follow-up construction of coding algorithms. To further consider recovery and security issues existing in the storage field, it also provides pluggable algorithmic patches based on the generated coding algorithms: path-based correcting and mapping shuffling. They provide approaches for probabilistic error correction and symmetric encryption respectively.

  Branch Decomposition-Independent Edit Distances for Merge Trees

• Authors: Florian Wetzels, Heike Leitte, Christoph Garth
• Subjects: Computational Geometry (cs.CG); Graphics (cs.GR)
• Arxiv link: https://arxiv.org/abs/2204.02919
• Pdf link: https://arxiv.org/pdf/2204.02919
• Abstract Edit distances between merge trees of scalar fields have many applications in scientific visualization, such as ensemble analysis, feature tracking or symmetry detection. In this paper, we propose branch mappings, a novel approach to the construction of edit mappings for merge trees. Classic edit mappings match nodes or edges of two trees onto each other, and therefore have to either rely on branch decompositions of both trees or have to use auxiliary node properties to determine a matching. In contrast, branch mappings employ branch properties instead of node similarity information, and are independent of predetermined branch decompositions. Especially for topological features, which are typically based on branch properties, this allows a more intuitive distance measure which is also less susceptible to instabilities from small-scale perturbations. We describe a quartic runtime algorithm for computing optimal branch mappings, which is faster than the only other branch decomposition-independent method in the literature by more than a linear factor. Furthermore, we compare the results of our method on synthetic and real-world examples to demonstrate its practicality and utility.

  "The Pedestrian next to the Lamppost" Adaptive Object Graphs for Better Instantaneous Mapping

• Authors: Avishkar Saha, Oscar Mendez, Chris Russell, Richard Bowden
• Subjects: Computer Vision and Pattern Recognition (cs.CV)
• Arxiv link: https://arxiv.org/abs/2204.02944
• Pdf link: https://arxiv.org/pdf/2204.02944
• Abstract Estimating a semantically segmented bird's-eye-view (BEV) map from a single image has become a popular technique for autonomous control and navigation. However, they show an increase in localization error with distance from the camera. While such an increase in error is entirely expected - localization is harder at distance - much of the drop in performance can be attributed to the cues used by current texture-based models, in particular, they make heavy use of object-ground intersections (such as shadows), which become increasingly sparse and uncertain for distant objects. In this work, we address these shortcomings in BEV-mapping by learning the spatial relationship between objects in a scene. We propose a graph neural network which predicts BEV objects from a monocular image by spatially reasoning about an object within the context of other objects. Our approach sets a new state-of-the-art in BEV estimation from monocular images across three large-scale datasets, including a 50% relative improvement for objects on nuScenes.

  Keyword: localization

  Robust Active Visual Perching with Quadrotors on Inclined Surfaces

• Authors: Jeffrey Mao, Stephen Nogar, Christopher Kroninger, Giuseppe Loianno
• Subjects: Robotics (cs.RO)
• Arxiv link: https://arxiv.org/abs/2204.02458
• Pdf link: https://arxiv.org/pdf/2204.02458
• Abstract Autonomous Micro Aerial Vehicles are deployed for a variety tasks including surveillance and monitoring. Perching and staring allow the vehicle to monitor targets without flying, saving battery power and increasing the overall mission time without the need to frequently replace batteries. This paper addresses the Active Visual Perching (AVP) control problem to autonomously perch on inclined surfaces up to $90^\circ$. Our approach generates dynamically feasible trajectories to navigate and perch on a desired target location, while taking into account actuator and Field of View (FoV) constraints. By replanning in mid-flight, we take advantage of more accurate target localization increasing the perching maneuver's robustness to target localization or control errors. We leverage the Karush-Kuhn-Tucker (KKT) conditions to identify the compatibility between planning objectives and the visual sensing constraint during the planned maneuver. Furthermore, we experimentally identify the corresponding boundary conditions that maximizes the spatio-temporal target visibility during the perching maneuver. The proposed approach works on-board in real-time with significant computational constraints relying exclusively on cameras and an Inertial Measurement Unit (IMU). Experimental results validate the proposed approach and shows the higher success rate as well as increased target interception precision and accuracy with respect to a one-shot planning approach, while still retaining aggressive capabilities with flight envelopes that include large excursions from the hover position on inclined surfaces up to 90$^\circ$, angular speeds up to 750~deg/s, and accelerations up to 10~m/s$^2$.

  SEAL: A Large-scale Video Dataset of Multi-grained Spatio-temporally Action Localization

• Authors: Shimin Chen, Wei Li, Chen Chen, Jianyang Gu, Jiaming Chu, Xunqiang Tao, Yandong Guo
• Subjects: Computer Vision and Pattern Recognition (cs.CV)
• Arxiv link: https://arxiv.org/abs/2204.02688
• Pdf link: https://arxiv.org/pdf/2204.02688
• Abstract In spite of many dataset efforts for human action recognition, current computer vision algorithms are still limited to coarse-grained spatial and temporal annotations among human daily life. In this paper, we introduce a novel large-scale video dataset dubbed SEAL for multi-grained Spatio-tEmporal Action Localization. SEAL consists of two kinds of annotations, SEAL Tubes and SEAL Clips. We observe that atomic actions can be combined into many complex activities. SEAL Tubes provide both atomic action and complex activity annotations in tubelet level, producing 49.6k atomic actions spanning 172 action categories and 17.7k complex activities spanning 200 activity categories. SEAL Clips localizes atomic actions in space during two-second clips, producing 510.4k action labels with multiple labels per person. Extensive experimental results show that SEAL significantly helps to advance video understanding.

  Exploration with Global Consistency Using Real-Time Re-integration and Active Loop Closure

• Authors: Yichen Zhang, Boyu Zhou, Luqi Wang, Shaojie Shen
• Subjects: Robotics (cs.RO)
• Arxiv link: https://arxiv.org/abs/2204.02693
• Pdf link: https://arxiv.org/pdf/2204.02693
• Abstract Despite recent progress of robotic exploration, most methods assume that drift-free localization is available, which is problematic in reality and causes severe distortion of the reconstructed map. In this work, we present a systematic exploration mapping and planning framework that deals with drifted localization, allowing efficient and globally consistent reconstruction. A real-time re-integration-based mapping approach along with a frame pruning mechanism is proposed, which rectifies map distortion effectively when drifted localization is corrected upon detecting loop-closure. Besides, an exploration planning method considering historical viewpoints is presented to enable active loop closing, which promotes a higher opportunity to correct localization errors and further improves the mapping quality. We evaluate both the mapping and planning methods as well as the entire system comprehensively in simulation and real-world experiments, showing their effectiveness in practice. The implementation of the proposed method will be made open-source for the benefit of the robotics community.

  Intervertebral Disc Labeling With Learning Shape Information, A Look Once Approach

• Authors: Reza Azad, Moein Heidari, Julien Cohen-Adad, Ehsan Adeli, Dorit Merhof
• Subjects: Computer Vision and Pattern Recognition (cs.CV)
• Arxiv link: https://arxiv.org/abs/2204.02943
• Pdf link: https://arxiv.org/pdf/2204.02943
• Abstract Accurate and automatic segmentation of intervertebral discs from medical images is a critical task for the assessment of spine-related diseases such as osteoporosis, vertebral fractures, and intervertebral disc herniation. To date, various approaches have been developed in the literature which routinely relies on detecting the discs as the primary step. A disadvantage of many cohort studies is that the localization algorithm also yields false-positive detections. In this study, we aim to alleviate this problem by proposing a novel U-Net-based structure to predict a set of candidates for intervertebral disc locations. In our design, we integrate the image shape information (image gradients) to encourage the model to learn rich and generic geometrical information. This additional signal guides the model to selectively emphasize the contextual representation and suppress the less discriminative features. On the post-processing side, to further decrease the false positive rate, we propose a permutation invariant 'look once' model, which accelerates the candidate recovery procedure. In comparison with previous studies, our proposed approach does not need to perform the selection in an iterative fashion. The proposed method was evaluated on the spine generic public multi-center dataset and demonstrated superior performance compared to previous work. We have provided the implementation code in https://github.com/rezazad68/intervertebral-lookonce

  "The Pedestrian next to the Lamppost" Adaptive Object Graphs for Better Instantaneous Mapping

• Authors: Avishkar Saha, Oscar Mendez, Chris Russell, Richard Bowden
• Subjects: Computer Vision and Pattern Recognition (cs.CV)
• Arxiv link: https://arxiv.org/abs/2204.02944
• Pdf link: https://arxiv.org/pdf/2204.02944
• Abstract Estimating a semantically segmented bird's-eye-view (BEV) map from a single image has become a popular technique for autonomous control and navigation. However, they show an increase in localization error with distance from the camera. While such an increase in error is entirely expected - localization is harder at distance - much of the drop in performance can be attributed to the cues used by current texture-based models, in particular, they make heavy use of object-ground intersections (such as shadows), which become increasingly sparse and uncertain for distant objects. In this work, we address these shortcomings in BEV-mapping by learning the spatial relationship between objects in a scene. We propose a graph neural network which predicts BEV objects from a monocular image by spatially reasoning about an object within the context of other objects. Our approach sets a new state-of-the-art in BEV estimation from monocular images across three large-scale datasets, including a 50% relative improvement for objects on nuScenes.

[HaolingZHANG]

Would you like to update the author list of SPIDER-WEB enables stable, repairable, and encryptible algorithms under arbitrary local biochemical constraints in DNA-based storage? That is Haoling Zhang, Zhaojun Lan, Wenwei Zhang, Xun Xu, Zhi Ping, Yiwei Zhang, Yue Shen.

[zhuhu00]

Would you like to update the author list of SPIDER-WEB enables stable, repairable, and encryptible algorithms under arbitrary local biochemical constraints in DNA-based storage? That is Haoling Zhang, Zhaojun Lan, Wenwei Zhang, Xun Xu, Zhi Ping, Yiwei Zhang, Yue Shen.

Thank you very much for your attention. I have revised it. The main purpose of this repository is to grab the data of papers on Arxiv through keywords.

[HaolingZHANG]

never mind, arXiv platform linked the author list incorrectly