We demonstrate how the often overlooked inherent properties of large-scale LiDAR point clouds can be effectively utilized for self-supervised representation learning. In pursuit of this goal, we design a highly data-efficient feature pre-training backbone that considerably reduces the need for tedious 3D annotations to train state-of-the-art object detectors. We propose Masked AutoEncoder for LiDAR point clouds (MAELi) that intuitively leverages the sparsity of LiDAR point clouds in both the encoder and decoder during reconstruction. Our approach results in more expressive and useful features, which can be directly applied to downstream perception tasks, such as 3D object detection for autonomous driving. In a novel reconstruction schema, MAELi distinguishes between free and occluded space and employs a new masking strategy that targets the LiDAR's inherent spherical projection. To demonstrate the potential of MAELi, we pre-train one of the most widely-used 3D backbones in an end-to-end manner and show the effectiveness of our unsupervised pre-trained features on various 3D object detection architectures. Our method achieves significant performance improvements when only a small fraction of labeled frames is available for fine-tuning object detectors. For instance, with ~800 labeled frames, MAELi features enhance a SECOND model by +10.79APH/LEVEL 2 on Waymo Vehicles.
相關內容
Language models (LMs) have introduced a major paradigm shift in Natural Language Processing (NLP) modeling where large pre-trained LMs became integral to most of the NLP tasks. The LMs are intelligent enough to find useful and relevant representations of the language without any supervision. Perhaps, these models are used to fine-tune typical NLP tasks with significantly high accuracy as compared to the traditional approaches. Conversely, the training of these models requires a massively large corpus that is a good representation of the language. English LMs generally perform better than their other language counterparts, due to the availability of massive English corpora. This work elaborates on the design and development of a large Arabic corpus. It consists of over 500 GB of Arabic cleaned text targeted at improving cross-domain knowledge and downstream generalization capability of large-scale language models. Moreover, the corpus is utilized in the training of a large Arabic LM. In order to evaluate the effectiveness of the LM, a number of typical NLP tasks are fine-tuned. The tasks demonstrate a significant boost from 4.5 to 8.5% when compared to tasks fine-tuned on multi-lingual BERT (mBERT). To the best of my knowledge, this is currently the largest clean and diverse Arabic corpus ever collected.
Point cloud data has been extensively studied due to its compact form and flexibility in representing complex 3D structures. The ability of point cloud data to accurately capture and represent intricate 3D geometry makes it an ideal choice for a wide range of applications, including computer vision, robotics, and autonomous driving, all of which require an understanding of the underlying spatial structures. Given the challenges associated with annotating large-scale point clouds, self-supervised point cloud representation learning has attracted increasing attention in recent years. This approach aims to learn generic and useful point cloud representations from unlabeled data, circumventing the need for extensive manual annotations. In this paper, we present a comprehensive survey of self-supervised point cloud representation learning using DNNs. We begin by presenting the motivation and general trends in recent research. We then briefly introduce the commonly used datasets and evaluation metrics. Following that, we delve into an extensive exploration of self-supervised point cloud representation learning methods based on these techniques. Finally, we share our thoughts on some of the challenges and potential issues that future research in self-supervised learning for pre-training 3D point clouds may encounter.
While some powerful neural network architectures (e.g., Transformer, Graph Neural Networks) have achieved improved performance in sequential recommendation with high-order item dependency modeling, they may suffer from poor representation capability in label scarcity scenarios. To address the issue of insufficient labels, Contrastive Learning (CL) has attracted much attention in recent methods to perform data augmentation through embedding contrasting for self-supervision. However, due to the hand-crafted property of their contrastive view generation strategies, existing CL-enhanced models i) can hardly yield consistent performance on diverse sequential recommendation tasks; ii) may not be immune to user behavior data noise. In light of this, we propose a simple yet effective graph masked autoencoder that adaptively and dynamically distills global item transitional information for self-supervised augmentation. It naturally avoids the above issue of heavy reliance on constructing high-quality embedding contrastive views. Instead, an adaptive data reconstruction paradigm is designed to be integrated with the long-range item dependency modeling, for informative augmentation in sequential recommendation. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art baseline models and can learn more accurate representations against data noise and sparsity. Our implemented model code is available at //github.com/HKUDS/GMRec.
We propose octree-based transformers, named OctFormer, for 3D point cloud learning. OctFormer can not only serve as a general and effective backbone for 3D point cloud segmentation and object detection but also have linear complexity and is scalable for large-scale point clouds. The key challenge in applying transformers to point clouds is reducing the quadratic, thus overwhelming, computation complexity of attentions. To combat this issue, several works divide point clouds into non-overlapping windows and constrain attentions in each local window. However, the point number in each window varies greatly, impeding the efficient execution on GPU. Observing that attentions are robust to the shapes of local windows, we propose a novel octree attention, which leverages sorted shuffled keys of octrees to partition point clouds into local windows containing a fixed number of points while permitting shapes of windows to change freely. And we also introduce dilated octree attention to expand the receptive field further. Our octree attention can be implemented in 10 lines of code with open-sourced libraries and runs 17 times faster than other point cloud attentions when the point number exceeds 200k. Built upon the octree attention, OctFormer can be easily scaled up and achieves state-of-the-art performances on a series of 3D segmentation and detection benchmarks, surpassing previous sparse-voxel-based CNNs and point cloud transformers in terms of both efficiency and effectiveness. Notably, on the challenging ScanNet200 dataset, OctFormer outperforms sparse-voxel-based CNNs by 7.3 in mIoU. Our code and trained models are available at //wang-ps.github.io/octformer.
Feature Descriptors and Detectors are two main components of feature-based point cloud registration. However, little attention has been drawn to the explicit representation of local and global semantics in the learning of descriptors and detectors. In this paper, we present a framework that explicitly extracts dual-level descriptors and detectors and performs coarse-to-fine matching with them. First, to explicitly learn local and global semantics, we propose a hierarchical contrastive learning strategy, training the robust matching ability of high-level descriptors, and refining the local feature space using low-level descriptors. Furthermore, we propose to learn dual-level saliency maps that extract two groups of keypoints in two different senses. To overcome the weak supervision of binary matchability labels, we propose a ranking strategy to label the significance ranking of keypoints, and thus provide more fine-grained supervision signals. Finally, we propose a global-to-local matching scheme to obtain robust and accurate correspondences by leveraging the complementary dual-level features.Quantitative experiments on 3DMatch and KITTI odometry datasets show that our method achieves robust and accurate point cloud registration and outperforms recent keypoint-based methods.
Most deepfake detection methods focus on detecting spatial and/or spatio-temporal changes in facial attributes. This is because available benchmark datasets contain mostly visual-only modifications. However, a sophisticated deepfake may include small segments of audio or audio-visual manipulations that can completely change the meaning of the content. To addresses this gap, we propose and benchmark a new dataset, Localized Audio Visual DeepFake (LAV-DF), consisting of strategic content-driven audio, visual and audio-visual manipulations. The proposed baseline method, Boundary Aware Temporal Forgery Detection (BA-TFD), is a 3D Convolutional Neural Network-based architecture which efficiently captures multimodal manipulations. We further improve (i.e. BA-TFD+) the baseline method by replacing the backbone with a Multiscale Vision Transformer and guide the training process with contrastive, frame classification, boundary matching and multimodal boundary matching loss functions. The quantitative analysis demonstrates the superiority of BA- TFD+ on temporal forgery localization and deepfake detection tasks using several benchmark datasets including our newly proposed dataset. The dataset, models and code are available at //github.com/ControlNet/LAV-DF.
We present the UrbanBIS benchmark for large-scale 3D urban understanding, supporting practical urban-level semantic and building-level instance segmentation. UrbanBIS comprises six real urban scenes, with 2.5 billion points, covering a vast area of 10.78 square kilometers and 3,370 buildings, captured by 113,346 views of aerial photogrammetry. Particularly, UrbanBIS provides not only semantic-level annotations on a rich set of urban objects, including buildings, vehicles, vegetation, roads, and bridges, but also instance-level annotations on the buildings. Further, UrbanBIS is the first 3D dataset that introduces fine-grained building sub-categories, considering a wide variety of shapes for different building types. Besides, we propose B-Seg, a building instance segmentation method to establish UrbanBIS. B-Seg adopts an end-to-end framework with a simple yet effective strategy for handling large-scale point clouds. Compared with mainstream methods, B-Seg achieves better accuracy with faster inference speed on UrbanBIS. In addition to the carefully-annotated point clouds, UrbanBIS provides high-resolution aerial-acquisition photos and high-quality large-scale 3D reconstruction models, which shall facilitate a wide range of studies such as multi-view stereo, urban LOD generation, aerial path planning, autonomous navigation, road network extraction, and so on, thus serving as an important platform for many intelligent city applications.
Future collaborative robots must be capable of finding objects. As such a fundamental skill, we expect object search to eventually become an off-the-shelf capability for any robot, similar to e.g., object detection, SLAM, and motion planning. However, existing approaches either make unrealistic compromises (e.g., reduce the problem from 3D to 2D), resort to ad-hoc, greedy search strategies, or attempt to learn end-to-end policies in simulation that are yet to generalize across real robots and environments. This thesis argues that through using Partially Observable Markov Decision Processes (POMDPs) to model object search while exploiting structures in the human world (e.g., octrees, correlations) and in human-robot interaction (e.g., spatial language), a practical and effective system for generalized object search can be achieved. In support of this argument, I develop methods and systems for (multi-)object search in 3D environments under uncertainty due to limited field of view, occlusion, noisy, unreliable detectors, spatial correlations between objects, and possibly ambiguous spatial language (e.g., "The red car is behind Chase Bank"). Besides evaluation in simulators such as PyGame, AirSim, and AI2-THOR, I design and implement a robot-independent, environment-agnostic system for generalized object search in 3D and deploy it on the Boston Dynamics Spot robot, the Kinova MOVO robot, and the Universal Robots UR5e robotic arm, to perform object search in different environments. The system enables, for example, a Spot robot to find a toy cat hidden underneath a couch in a kitchen area in under one minute. This thesis also broadly surveys the object search literature, proposing taxonomies in object search problem settings, methods and systems.
Masked autoencoders are scalable vision learners, as the title of MAE \cite{he2022masked}, which suggests that self-supervised learning (SSL) in vision might undertake a similar trajectory as in NLP. Specifically, generative pretext tasks with the masked prediction (e.g., BERT) have become a de facto standard SSL practice in NLP. By contrast, early attempts at generative methods in vision have been buried by their discriminative counterparts (like contrastive learning); however, the success of mask image modeling has revived the masking autoencoder (often termed denoising autoencoder in the past). As a milestone to bridge the gap with BERT in NLP, masked autoencoder has attracted unprecedented attention for SSL in vision and beyond. This work conducts a comprehensive survey of masked autoencoders to shed insight on a promising direction of SSL. As the first to review SSL with masked autoencoders, this work focuses on its application in vision by discussing its historical developments, recent progress, and implications for diverse applications.
This paper shows that masked autoencoders (MAE) are scalable self-supervised learners for computer vision. Our MAE approach is simple: we mask random patches of the input image and reconstruct the missing pixels. It is based on two core designs. First, we develop an asymmetric encoder-decoder architecture, with an encoder that operates only on the visible subset of patches (without mask tokens), along with a lightweight decoder that reconstructs the original image from the latent representation and mask tokens. Second, we find that masking a high proportion of the input image, e.g., 75%, yields a nontrivial and meaningful self-supervisory task. Coupling these two designs enables us to train large models efficiently and effectively: we accelerate training (by 3x or more) and improve accuracy. Our scalable approach allows for learning high-capacity models that generalize well: e.g., a vanilla ViT-Huge model achieves the best accuracy (87.8%) among methods that use only ImageNet-1K data. Transfer performance in downstream tasks outperforms supervised pre-training and shows promising scaling behavior.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191