Kitti
PulseAugur coverage of Kitti — every cluster mentioning Kitti across labs, papers, and developer communities, ranked by signal.
4 天有情绪数据
-
新方法利用证据图精炼激光雷达-相机标定
研究人员开发了一种新的激光雷达和相机系统标定方法，特别适用于农业环境。该方法采用“支持图驱动”技术来识别对准确标定最关键的观测，过滤掉噪声或模糊的数据。通过聚合对齐观测的一致性，该方法突出了可靠的标定证据，提高了在KITTI等数据集上的准确性。
-
新数据集应对洪水路况下的自动驾驶挑战
研究人员推出了洪水路况数据集（FRED），这是首个专为洪水条件下自动驾驶设计的全模态数据集。FRED 包含来自摄像头、LiDAR 和 IMU 传感器的同步数据，在洪水事件期间及之后在五个地点采集。该数据集以 KITTI 风格和 RTMaps 格式发布，并附带语义标签，以促进水患检测系统的开发和评估。
-
MonoPRIO advances monocular 3D object detection with adaptive priors
Researchers have introduced MonoPRIO, a novel approach to monocular 3D object detection that addresses the challenge of accurately determining object size and depth from a single image. The method employs adaptive prior…
-
CoopNet improves self-supervised depth, odometry, and optical flow predictions
Researchers have developed CoopNet, a novel method to enhance self-supervised learning for predicting depth, odometry, and optical flow. This approach dynamically adjusts gradient apportionment to ensure balanced learni…
-
Conformal prediction enhances object detection uncertainty
Researchers have developed a new method for probabilistic object detection using conformal prediction, enhancing uncertainty quantification for safety-critical applications like autonomous driving. This approach adapts …
-
New AI framework proactively monitors ADAS camera reliability before failure
Researchers have developed a new framework for monitoring the reliability of cameras used in Advanced Driver-Assistance Systems (ADAS). This system proactively estimates perception risk by analyzing degradation-induced …
-
BEVCALIB model uses bird's-eye view features for LiDAR-camera calibration
Researchers have developed BEVCALIB, a novel method for calibrating LiDAR and camera sensors, crucial for autonomous driving systems. This approach utilizes bird's-eye view (BEV) features extracted from both sensor type…
-
Pixel-to-4D generates videos from single images with dynamic 3D Gaussians
Researchers have developed a new framework for generating videos from single images, offering enhanced control over camera movement. This method utilizes dynamic 3D Gaussian representations to model scenes and sample ob…
-
TACO管线融合IMU和跨视图地理定位以实现精确导航
研究人员开发了TACO，这是一个新管线，它将惯性测量单元（IMU）数据与细粒度的跨视图地理定位（CVGL）紧密集成，可在没有连续GNSS信号的情况下实现精确的定位。该系统旨在在GNSS不可靠的环境中提供准确的位置修复，例如城市峡谷或信号干扰区域。TACO在KITTI数据集上展示了绝对轨迹误差的显著降低，从97.0米提高到16.3米，同时保持了较低的计算成本。
-
LoGeR model enables long-context geometric reconstruction with hybrid memory
Researchers have introduced LoGeR, a novel architecture designed for long-context geometric reconstruction in videos. This system addresses the limitations of existing feedforward models by processing video streams in c…
-
OpenVO framework enhances visual odometry with temporal awareness and geometric priors
Researchers have developed OpenVO, a new framework for open-world visual odometry that accounts for temporal dynamics and works with uncalibrated cameras. Unlike previous methods that assume fixed observation frequencie…
-
通过具有可学习代数群和环结构的神经网络进行单目深度估计
研究人员开发了LAGRNet，这是一个用于单目深度估计的新框架，它融入了代数几何原理。与先前将深度估计视为通用回归问题的先前方法不同，LAGRNet明确地将可学习的群、环和层结构嵌入其深度学习流程中。这种方法旨在强制执行投影等变性并确保全局拓扑一致性，从而在KITTI和NYU-Depth V2等基准测试中提高准确性和泛化能力。
-
PoseFM uses Flow Matching for robust monocular visual odometry
Researchers have introduced PoseFM, a novel framework that reframes monocular visual odometry as a generative task using Flow Matching. This approach models camera motion as a distribution, allowing for uncertainty esti…