Search

The scarcity and complexity of voxel-level annotations in 3D medical imaging present significant challenges, particularly due to the domain gap between labeled datasets from well-resourced centers and unlabeled datasets from less-resourced centers. This disparity affects the fairness of artificial intelligence algorithms in healthcare. We introduce Diffuse-UDA, a novel method leveraging diffusion models to tackle Unsupervised Domain Adaptation (UDA) in medical image segmentation. Diffuse-UDA generates high-quality image-mask pairs with target domain characteristics and various structures, thereby enhancing UDA tasks. Initially, pseudo labels for target domain samples are generated. Subsequently, a specially tailored diffusion model, incorporating deformable augmentations, is trained on image-label or image-pseudo-label pairs from both domains. Finally, source domain labels guide the diffusion model to generate image-label pairs for the target domain. Comprehensive evaluations on several benchmarks demonstrate that Diffuse-UDA outperforms leading UDA and semi-supervised strategies, achieving performance close to or even surpassing the theoretical upper bound of models trained directly on target domain data. Diffuse-UDA offers a pathway to advance the development and deployment of AI systems in medical imaging, addressing disparities between healthcare environments. This approach enables the exploration of innovative AI-driven diagnostic tools, improves outcomes, saves time, and reduces human error.

Star-forming regions, characterized by dense environments, experience frequent encounters that significantly influence binary systems, leading to their hardening, softening, or ionization. We extend the Hut \& Bahcall formalism to derive an analytical expression for the ionization cross-section in extreme mass ratio binary systems, allowing us to investigate the orbital evolution and population dynamics of binary planets and binary brown dwarfs in star clusters, while considering ongoing binary system formation. Our findings reveal that for low-mass soft binaries, the semi-major axis distribution asymptotes to a universal power law between $\propto a^{-8/3}$ and $\propto a^{-5/3}$ over the derived ionization timescale. We also discuss the implications of our results for the candidate Jupiter-mass binary objects putatively reported in the Trapezium cluster. We demonstrate that if their existence is verified, they likely form continuously with a spectrum proportional to $a^{1}$, aligning better with the ejection mechanism than with the in-situ formation mechanism, which predicts a distribution roughly proportional to $a^{-1}$. However, this implies an impractically high ejection formation rate. Alternatively, if these objects are binary brown dwarfs, continuous in-situ formation ($\propto a^{-1}$) with an initial minimal semi-major axis around 20 AU and a formation rate of 100 Myr$^{-1}$ plausibly matches the observed number of single objects, binary number, binary fraction, and semi-major axis distribution., Comment: Comments are welcome

We study area-minimizing hypersurfaces in singular ambient manifolds whose tangent cones have nonnegative scalar curvature on their regular parts. We prove that the singular set of the hypersurface has codimension at least 3 in our settings. We also give an example in which n-dimensional minimizing hypersurface in an ambient space as given above may have singularities of dimension $n-3$. This shows that our regularity result is sharp., Comment: 34 pages. Comments are welcome

The phenomenon of changing-look (CL) behavior in active galactic nuclei (AGN) is characterized by dramatic changes in luminosity and/or emission line profiles over relatively short periods, ranging from months to years. The origin of CL-AGNs remains a mystery, but one proposed explanation involves the response of the inner AGN disk to tidal disruption events (TDEs) around the supermassive black hole (SMBH). In this Letter, we calculate the predicted frequency of AGN TDEs as a function of SMBH mass and compare the results to the observed CL-AGN distribution. We find that if the fraction of CL-AGNs caused by AGN-TDEs is high, then: (1) most SMBHs in CL-AGN are near maximal spin, with the dimensionless spin parameter $a>0.9$; (2) AGN inner disks have a high surface density ($\geq 10^{7}\, {\rm g\, cm^{-2}}$); (3) typical AGN lifetimes are $\sim 10$-$100$ Myr; and (4) a nuclear star cluster initial mass function (IMF) that scales as $\sim m_*^{-1.6}$ is preferred. Future observations of CL-AGN will help constrain the fraction of CL-AGNs caused by AGN-TDEs, SMBH spins, AGN lifetimes, and the nuclear star cluster IMF., Comment: Comments are welcome

This study evaluates the effectiveness of zero-shot compression techniques on large language models (LLMs) under long-context. We identify the tendency for computational errors to increase under long-context when employing certain compression methods. We propose a hypothesis to explain the varied behavior of different LLM compression techniques and explore remedies to mitigate the performance decline observed in some techniques under long-context. This is a course report for COS 598D Machine Learning and Systems by Prof. Kai Li at Princeton University. Due to limited computational resources, our experiments were conducted only on LLaMA-2-7B-32K.

Machine unlearning provides viable solutions to revoke the effect of certain training data on pre-trained model parameters. Existing approaches provide unlearning recipes for classification and generative models. However, a category of important machine learning models, i.e., contrastive learning (CL) methods, is overlooked. In this paper, we fill this gap by first proposing the framework of Machine Unlearning for Contrastive learning (MUC) and adapting existing methods. Furthermore, we observe that several methods are mediocre unlearners and existing auditing tools may not be sufficient for data owners to validate the unlearning effects in contrastive learning. We thus propose a novel method called Alignment Calibration (AC) by explicitly considering the properties of contrastive learning and optimizing towards novel auditing metrics to easily verify unlearning. We empirically compare AC with baseline methods on SimCLR, MoCo and CLIP. We observe that AC addresses drawbacks of existing methods: (1) achieving state-of-the-art performance and approximating exact unlearning (retraining); (2) allowing data owners to clearly visualize the effect caused by unlearning through black-box auditing.

We introduce SEA-RAFT, a more simple, efficient, and accurate RAFT for optical flow. Compared with RAFT, SEA-RAFT is trained with a new loss (mixture of Laplace). It directly regresses an initial flow for faster convergence in iterative refinements and introduces rigid-motion pre-training to improve generalization. SEA-RAFT achieves state-of-the-art accuracy on the Spring benchmark with a 3.69 endpoint-error (EPE) and a 0.36 1-pixel outlier rate (1px), representing 22.9% and 17.8% error reduction from best published results. In addition, SEA-RAFT obtains the best cross-dataset generalization on KITTI and Spring. With its high efficiency, SEA-RAFT operates at least 2.3x faster than existing methods while maintaining competitive performance. The code is publicly available at https://github.com/princeton-vl/SEA-RAFT.

Obtaining reliable precipitation estimation with high resolutions in time and space is of great importance to hydrological studies. However, accurately estimating precipitation is a challenging task over high mountainous complex terrain. The three widely used precipitation measurement approaches, namely rainfall gauge, precipitation radars, and satellite-based precipitation sensors, have their own pros and cons in producing reliable precipitation products over complex areas. One way to decrease the detection error probability and improve data reliability is precipitation data merging. With the rapid advancements in computational capabilities and the escalating volume and diversity of earth observational data, Deep Learning (DL) models have gained considerable attention in geoscience. In this study, a deep learning technique, namely Long Short-term Memory (LSTM), was employed to merge a radar-based and a satellite-based Global Precipitation Measurement (GPM) precipitation product Integrated Multi-Satellite Retrievals for GPM (IMERG) precipitation product at hourly scale. The merged results are compared with the widely used reanalysis precipitation product, Multi-Radar Multi-Sensor (MRMS), and assessed against gauge observational data from the California Data Exchange Center (CDEC). The findings indicated that the LSTM-based merged precipitation notably underestimated gauge observations and, at times, failed to provide meaningful estimates, showing predominantly near-zero values. Relying solely on individual Quantitative Precipitation Estimates (QPEs) without additional meteorological input proved insufficient for generating reliable merged QPE. However, the merged results effectively captured the temporal trends of the observations, outperforming MRMS in this aspect. This suggested that incorporating bias correction techniques could potentially enhance the accuracy of the merged product.

The origin of supermassive black holes (SMBHs) residing in the centers of most galaxies remains a mystery. The Event Horizon Telescope (EHT) provided direct imaging of the SMBH Sgr A* at the Milky Way's center, indicating it likely spins rapidly with its spin axis significantly misaligned relative to the Galactic plane's angular momentum. Through investigating various SMBH growth models, here we show that the inferred spin properties of Sgr A* provide evidence of a past SMBH merger. Inspired by the merger between the Milky Way and Gaia-Enceladus, which has a 4:1 mass ratio as inferred from Gaia data, we have discovered that a 4:1 major merger of SMBH with a binary angular momentum inclination angle of 145-180 degrees with respect to the line of sight (LOS), can successfully replicate the measured spin properties of Sgr A*. This merger event in our galaxy provides potential observational support for the theory of hierarchical BH mergers in the formation and growth of SMBHs. The inferred merger rate, consistent with theoretical predictions, suggests a promising detection rate of SMBH mergers for space-borne gravitational wave detectors expected to operate in 2030s., Comment: To appear in Nature Astronomy, authors' version

Long Time Coherent Integration (LTCI) aims to accumulate target energy through long time integration, which is an effective method for the detection of a weak target. However, for a moving target, defocusing can occur due to range migration (RM) and Doppler frequency migration (DFM). To address this issue, RM and DFM corrections are required in order to achieve a well-focused image for the subsequent detection. Since RM and DFM are induced by the same motion parameters, existing approaches such as the generalized Radon-Fourier transform (GRFT) or the keystone transform (KT)-matching filter process (MFP) adopt the same search space for the motion parameters in order to eliminate both effects, thus leading to large redundancy in computation. To this end, this paper first proposes a dual-scale decomposition of the target motion parameters, consisting of well designed coarse and fine motion parameters. Then, utilizing this decomposition, the joint correction of the RM and DFM effects is decoupled into a cascade procedure, first RM correction on the coarse search space and then DFM correction on the fine search spaces. As such, step size of the search space can be tailored to RM and DFM corrections, respectively, thus avoiding large redundant computation effectively. The resulting algorithms are called dual-scale GRFT (DS-GRFT) or dual-scale GRFT (DS-KTMFP) which provide comparable performance while achieving significant improvement in computational efficiency compared to standard GRFT (KT-MFP). Simulation experiments verify their effectiveness and efficiency.

To enable autonomous vehicles to perform discretionary lane change amidst the random traffic flow on highways, this paper introduces a decision-making and control method for vehicle lane change based on Model Predictive Control (MPC). This approach divides the driving control of vehicles on highways into two parts: lane-change decision and lane-change control, both of which are solved using the MPC method. In the lanechange decision module, the minimum driving costs for each lane are computed and compared by solving the MPC problem to make lane-change decisions. In the lane-change control module, a dynamic bicycle model is incorporated, and a multi-objective cost function is designed to obtain the optimal control inputs for the lane-change process. Additionally, A long-short term memory (LSTM) model is used to predict the trajectories of surrounding vehicles for both the MPC decision and control modules. The proposed lane-change decision and control method is simulated and validated in a driving simulator under random highway traffic conditions.

Although many large language models (LLMs) have been trained to refuse harmful requests, they are still vulnerable to jailbreaking attacks which rewrite the original prompt to conceal its harmful intent. In this paper, we propose a new method for defending LLMs against jailbreaking attacks by ``backtranslation''. Specifically, given an initial response generated by the target LLM from an input prompt, our backtranslation prompts a language model to infer an input prompt that can lead to the response. The inferred prompt is called the backtranslated prompt which tends to reveal the actual intent of the original prompt, since it is generated based on the LLM's response and not directly manipulated by the attacker. We then run the target LLM again on the backtranslated prompt, and we refuse the original prompt if the model refuses the backtranslated prompt. We explain that the proposed defense provides several benefits on its effectiveness and efficiency. We empirically demonstrate that our defense significantly outperforms the baselines, in the cases that are hard for the baselines, and our defense also has little impact on the generation quality for benign input prompts. Our implementation is based on our library for LLM jailbreaking defense algorithms at \url{https://github.com/YihanWang617/llm-jailbreaking-defense}, and the code for reproducing our experiments is available at \url{https://github.com/YihanWang617/LLM-Jailbreaking-Defense-Backtranslation}.

Availability attacks can prevent the unauthorized use of private data and commercial datasets by generating imperceptible noise and making unlearnable examples before release. Ideally, the obtained unlearnability prevents algorithms from training usable models. When supervised learning (SL) algorithms have failed, a malicious data collector possibly resorts to contrastive learning (CL) algorithms to bypass the protection. Through evaluation, we have found that most of the existing methods are unable to achieve both supervised and contrastive unlearnability, which poses risks to data protection. Different from recent methods based on contrastive error minimization, we employ contrastive-like data augmentations in supervised error minimization or maximization frameworks to obtain attacks effective for both SL and CL. Our proposed AUE and AAP attacks achieve state-of-the-art worst-case unlearnability across SL and CL algorithms with less computation consumption, showcasing prospects in real-world applications.

Unlearnable example attacks are data poisoning attacks aiming to degrade the clean test accuracy of deep learning by adding imperceptible perturbations to the training samples, which can be formulated as a bi-level optimization problem. However, directly solving this optimization problem is intractable for deep neural networks. In this paper, we investigate unlearnable example attacks from a game-theoretic perspective, by formulating the attack as a nonzero sum Stackelberg game. First, the existence of game equilibria is proved under the normal setting and the adversarial training setting. It is shown that the game equilibrium gives the most powerful poison attack in that the victim has the lowest test accuracy among all networks within the same hypothesis space, when certain loss functions are used. Second, we propose a novel attack method, called the Game Unlearnable Example (GUE), which has three main gradients. (1) The poisons are obtained by directly solving the equilibrium of the Stackelberg game with a first-order algorithm. (2) We employ an autoencoder-like generative network model as the poison attacker. (3) A novel payoff function is introduced to evaluate the performance of the poison. Comprehensive experiments demonstrate that GUE can effectively poison the model in various scenarios. Furthermore, the GUE still works by using a relatively small percentage of the training data to train the generator, and the poison generator can generalize to unseen data well. Our implementation code can be found at https://github.com/hong-xian/gue.

Stability analysis is an essential aspect of studying the generalization ability of deep learning, as it involves deriving generalization bounds for stochastic gradient descent-based training algorithms. Adversarial training is the most widely used defense against adversarial example attacks. However, previous generalization bounds for adversarial training have not included information regarding the data distribution. In this paper, we fill this gap by providing generalization bounds for stochastic gradient descent-based adversarial training that incorporate data distribution information. We utilize the concepts of on-average stability and high-order approximate Lipschitz conditions to examine how changes in data distribution and adversarial budget can affect robust generalization gaps. Our derived generalization bounds for both convex and non-convex losses are at least as good as the uniform stability-based counterparts which do not include data distribution information. Furthermore, our findings demonstrate how distribution shifts from data poisoning attacks can impact robust generalization.

The strong general capabilities of Large Language Models (LLMs) bring potential ethical risks if they are unrestrictedly accessible to malicious users. Token-level watermarking inserts watermarks in the generated texts by altering the token probability distributions with a private random number generator seeded by its prefix tokens. However, this watermarking algorithm alters the logits during generation, which can lead to a downgraded text quality if it chooses to promote tokens that are less relevant given the input. In this work, we propose to improve the quality of texts generated by a watermarked language model by Watermarking with Importance Scoring (WIS). At each generation step, we estimate the importance of the token to generate, and prevent it from being impacted by watermarking if it is important for the semantic correctness of the output. We further propose three methods to predict importance scoring, including a perturbation-based method and two model-based methods. Empirical experiments show that our method can generate texts with better quality with comparable level of detection rate., Comment: Work in progress

4D radar is recognized for its resilience and cost-effectiveness under adverse weather conditions, thus playing a pivotal role in autonomous driving. While cameras and LiDAR are typically the primary sensors used in perception modules for autonomous vehicles, radar serves as a valuable supplementary sensor. Unlike LiDAR and cameras, radar remains unimpaired by harsh weather conditions, thereby offering a dependable alternative in challenging environments. Developing radar-based 3D object detection not only augments the competency of autonomous vehicles but also provides economic benefits. In response, we propose the Multi-View Feature Assisted Network (\textit{MVFAN}), an end-to-end, anchor-free, and single-stage framework for 4D-radar-based 3D object detection for autonomous vehicles. We tackle the issue of insufficient feature utilization by introducing a novel Position Map Generation module to enhance feature learning by reweighing foreground and background points, and their features, considering the irregular distribution of radar point clouds. Additionally, we propose a pioneering backbone, the Radar Feature Assisted backbone, explicitly crafted to fully exploit the valuable Doppler velocity and reflectivity data provided by the 4D radar sensor. Comprehensive experiments and ablation studies carried out on Astyx and VoD datasets attest to the efficacy of our framework. The incorporation of Doppler velocity and RCS reflectivity dramatically improves the detection performance for small moving objects such as pedestrians and cyclists. Consequently, our approach culminates in a highly optimized 4D-radar-based 3D object detection capability for autonomous driving systems, setting a new standard in the field., Comment: 19 Pages, 7 figures, Accepted by ICONIP 2023

The discovery of planetary systems beyond our solar system has challenged established theories of planetary formation. Planetary orbits display a variety of unexpected architectures, and free-floating planets appear ubiquitous. The recent detection of candidate Jupiter Mass Binary Objects (JuMBOs) by the James Webb Space Telescope (JWST) has added another puzzling layer. Here, through direct few-body simulations, we demonstrate that JuMBOs could arise from the ejection of double giant planets following a close encounter with a passing star, if the two planets are nearly aligned at the closest approach. These ejected JuMBOs typically possess an average semi-major axis approximately three times the orbital separation within their original planetary system and a high eccentricity, characterized by a superthermal distribution that sets them apart from those formed primordially. We estimate the JuMBO formation rate per planetary system in typical and densely populated clusters, revealing a significant environmental dependence. In dense clusters, this formation rate can reach a few percent for wide planetary systems. Comparative analysis of JuMBO rates and properties with current and forthcoming JWST observations across various environments promises insights into the conditions under which these giant planets formed in protoplanetary disks, thereby imposing constraints on theories of giant planet formation., Comment: Matches version published in Nature Astronomy; Videos are available at https://yihanwangastro.github.io/#posts2

Recent observations of changing-look active galactic nuclei (AGN) hint at a frequency of accretion activity not fully explained by tidal disruption events (TDEs) stemming from relaxation processes in nucleus star clusters (NSCs), traditionally estimated to occur at rates of $10^{-4}$ to $10^{-5}$ yr$^{-1}$ per galaxy. In this letter, we propose an enhanced TDE rate through the AGN disk capture process, presenting a viable explanation for the frequent transitions observed in changing-look AGN. Specifically, we investigate the interaction between the accretion disk and retrograde stars within NSCs, resulting in the rapid occurrence of TDEs within a condensed time frame. Through detailed calculations, we derive the time-dependent TDE rates for both relaxation-induced TDE and disk-captured TDE. Our analysis reveals that TDEs triggered by the disk capture process can notably amplify the TDE rate by several orders of magnitude during the AGN phase. This mechanism offers a potential explanation for the enhanced high-energy variability characteristic of changing-look AGNs., Comment: Submitted to ApJL. Comments are welcome

Tidal Disruption Events (TDEs) are routinely observed in quiescent galaxies, as stars from the nuclear star cluster are scattered into the loss cone of the central supermassive black hole (SMBH). TDEs are also expected to occur in Active Galactic Nuclei (AGN), due to scattering or orbital eccentricity pumping of stars embedded in the innermost regions of the AGN accretion disk. Encounters with embedded stellar-mass black holes (BH) can result in AGN $\mu$TDEs. AGN TDEs and $\mu$TDEs could therefore account for a fraction of observed AGN variability. Here, by performing scattering experiments with the few-body code {\tt SpaceHub}, we compute the probability of AGN TDEs and $\mu$TDEs as a result of 3-body interactions between stars and binary BHs. We find that AGN TDEs are more probable during the early life of the AGNs, when rates are $\sim (6\times 10^{-5}-5 \times 10^{-2}) (f_\bullet/0.01)$ $~\rm{AGN}^{-1}$~yr$^{-1}$ (where $f_\bullet$ is the ratio between the number density of BHs and stars), generally higher than in quiescent galactic nuclei. By contrast, $\mu$TDEs should occur throughout the AGN lifetime at a rate of $\sim (1\times 10^{-4} - 4\times 10^{-2} (f_\bullet/0.01)$ $~\rm{AGN}^{-1}$~yr$^{-1}$. Detection and characterization of AGN TDEs and $\mu$ AGN TDEs with future surveys using {\em Rubin} and {\em Roman} will help constrain the populations of stars and compact objects embedded in AGN disks, a key input for the LVK AGN channel., Comment: Final version accepted by MNRAS

A long-lived magnetar, potentially originating from a binary neutron star system, has been proposed to explain the extended emission observed in certain short-duration gamma-ray bursts (sGRBs), and is posited as a potential central engine to power the engine-fed kilonovae. Previously, the process by which energy is injected into the surrounding ejecta/jet was widely believed to be nearly isotropic. In this study, we employ special relativity magnetohydrodynamic (SRMHD) simulations to investigate the wind injection process from a magnetar central engine. We explore the dynamics and energy distribution within the system and found that the parameter $\alpha=u_{\rm A}/u_{\rm MWN}$ can be used to indicate the collimation of the magnetar wind energy injection, where $u_{\rm A}$ is the local Alfven four-speed and $u_{\rm MWN}$ is the four-speed of the magnetar wind nebular (MWN) formed from wind-ejecta collision. A significant portion of the injected energy from the magnetar spin-down wind will be channeled to the jet axis due to collimation within the MWN. Achieving isotropic energy injection requires a significantly small $\alpha$ that necessitates either an ultra-relativistic expanding MWN or an extremely low magnetization MWN, both of which are challenging to attain in sGRBs. Consequently, a considerably reduced energy budget (i.e. energy per solid angle reduced by a factor of up to 10 with respect to the value under isotropic assumption) is anticipated to be injected into the ejecta for engine-fed kilonovae. Engine-fed kilonovae would appear fainter than originally anticipated., Comment: Accepted for publication in MNRAS

Robust 3D object detection in extreme weather and illumination conditions is a challenging task. While radars and thermal cameras are known for their resilience to these conditions, few studies have been conducted on radar-thermal fusion due to the lack of corresponding datasets. To address this gap, we first present a new multi-modal dataset called ThermRad, which includes a 3D LiDAR, a 4D radar, an RGB camera and a thermal camera. This dataset is unique because it includes data from all four sensors in extreme weather conditions, providing a valuable resource for future research in this area. To validate the robustness of 4D radars and thermal cameras for 3D object detection in challenging weather conditions, we propose a new multi-modal fusion method called RTDF-RCNN, which leverages the complementary strengths of 4D radars and thermal cameras to boost object detection performance. To further prove the effectiveness of our proposed framework, we re-implement state-of-the-art (SOTA) 3D detectors on our dataset as benchmarks for evaluation. Our method achieves significant enhancements in detecting cars, pedestrians, and cyclists, with improvements of over 7.98%, 24.27%, and 27.15%, respectively, while achieving comparable results to LiDAR-based approaches. Our contributions in both the ThermRad dataset and the new multi-modal fusion method provide a new approach to robust 3D object detection in adverse weather and illumination conditions. The ThermRad dataset will be released., Comment: At this time, we have not reached a definitive agreement regarding the ownership and copyright of this dataset. Due to the unresolved issue regarding the dataset, I am writing to formally request the withdrawal of our paper

The Active Galatic Nuclei(AGN) disk has been proposed as a potential channel for the merger of binary black holes. The population of massive stars and black holes in AGN disks captured from the nuclei cluster plays a crucial role in determining the efficiency of binary formation and final merger rate within the AGN disks. In this paper, we investigate the capture process using analytical and numerical approaches. We discover a new constant integral of motion for one object's capture process. Applying this result to the whole population of the nuclei cluster captured by the AGN disk, we find that the population of captured objects depends on the angular density and eccentricity distribution of the nuclei clusters and is effectively independent of the radial density profile of the nuclei cluster and disk models. An isotropic nuclei cluster with thermal eccentricity distribution predicts a captured profile $d N/d r \propto r^{-1/4}$. The captured objects are found to be dynamically crowded within the disk. Direct binary formation right after the capture would be promising, especially for stars. The conventional migration traps that help pile up single objects in AGN disks for black hole mergers might not be required., Comment: Typo corrections

LiDAR-based 3D object detection is of paramount importance for autonomous driving. Recent trends show a remarkable improvement for bird's-eye-view (BEV) based and point-based methods as they demonstrate superior performance compared to range-view counterparts. This paper presents an insight that leverages range-view representation to enhance 3D points for accurate 3D object detection. Specifically, we introduce a Redemption from Range-view Module (R2M), a plug-and-play approach for 3D surface texture enhancement from the 2D range view to the 3D point view. R2M comprises BasicBlock for 2D feature extraction, Hierarchical-dilated (HD) Meta Kernel for expanding the 3D receptive field, and Feature Points Redemption (FPR) for recovering 3D surface texture information. R2M can be seamlessly integrated into state-of-the-art LiDAR-based 3D object detectors as preprocessing and achieve appealing improvement, e.g., 1.39%, 1.67%, and 1.97% mAP improvement on easy, moderate, and hard difficulty level of KITTI val set, respectively. Based on R2M, we further propose R2Detector (R2Det) with the Synchronous-Grid RoI Pooling for accurate box refinement. R2Det outperforms existing range-view-based methods by a significant margin on both the KITTI benchmark and the Waymo Open Dataset. Codes will be made publicly available.

Invariance to spatial transformations such as translations and rotations is a desirable property and a basic design principle for classification neural networks. However, the commonly used convolutional neural networks (CNNs) are actually very sensitive to even small translations. There exist vast works to achieve exact or approximate transformation invariance by designing transformation-invariant models or assessing the transformations. These works usually make changes to the standard CNNs and harm the performance on standard datasets. In this paper, rather than modifying the classifier, we propose a pre-classifier restorer to recover translated (or even rotated) inputs to the original ones which will be fed into any classifier for the same dataset. The restorer is based on a theoretical result which gives a sufficient and necessary condition for an affine operator to be translational equivariant on a tensor space.

We study a system of intervals $I_1,\ldots,I_k$ on the real line and a continuous map $f$ with $f(I_1 \cup I_2 \cup \ldots \cup I_k)\supseteq I_1 \cup I_2 \cup \ldots \cup I_k$. It's conjectured that there exists a periodic point of period $\le k$ in $I_1\cup \ldots \cup I_k$. In this paper, we prove the conjecture by a discretization method and reduce the initial problem to an interesting combinatorial lemma concerning cyclic permutations. We also obtain a non-concentration property of periodic points of small periods in intervals., Comment: 12 pages, 3 figures. arXiv admin note: substantial text overlap with arXiv:2205.07225

We introduce Infinigen, a procedural generator of photorealistic 3D scenes of the natural world. Infinigen is entirely procedural: every asset, from shape to texture, is generated from scratch via randomized mathematical rules, using no external source and allowing infinite variation and composition. Infinigen offers broad coverage of objects and scenes in the natural world including plants, animals, terrains, and natural phenomena such as fire, cloud, rain, and snow. Infinigen can be used to generate unlimited, diverse training data for a wide range of computer vision tasks including object detection, semantic segmentation, optical flow, and 3D reconstruction. We expect Infinigen to be a useful resource for computer vision research and beyond. Please visit https://infinigen.org for videos, code and pre-generated data., Comment: Accepted to CVPR 2023, Camera Ready Version. Update 06/26/23: Change the open-source license to BSD