This repository collects code, videos related to point cloud transfer learning
[A Survey on Deep Domain Adaptation for LiDAR Perception]
[Deep Learning for 3D Point Clouds: A Survey]
[Deep Learning for LiDAR Point Clouds in Autonomous Driving: A Review]
[Unsupervised Point Cloud Representation Learning with Deep Neural Networks: A Survey]
[Deep Learning for LiDAR-Based Autonomous Vehicles in Smart Cities]
Self-supervised learning (SSL) allows to learn useful representations from unlabeled data and has been applied effectively for domain adaptation (DA) on images. It is still unknown if and how it can be leveraged for domain adaptation for 3D perception. Here we describe the first study of SSL for DA on point clouds. We introduce a new family of pretext tasks, Deformation Reconstruction, motivated by the deformations encountered in sim-to-real transformations. The key idea is to deform regions of the input shape and use a neural network to reconstruct them. We design three types of shape deformation methods: (1) Volume-based: shape deformation based on proximity in the input space; (2) Feature-based: deforming regions in the shape that are semantically similar; and (3) Sampling-based: shape deformation based on three simple sampling schemes. As a separate contribution, we also develop a new method based on the Mixup training procedure for point-clouds. Evaluations on six domain adaptations across synthetic and real furniture data, demonstrate large improvement over previous work. [Paper] [code]
A considerable amount of annotated training data is necessary to achieve state-of-the-art performance in perception tasks using point clouds. Unlike RGB-images, LiDAR point clouds captured with different sensors or varied mounting positions exhibit a significant shift in their input data distribution. This can impede transfer of trained feature extractors between datasets as it degrades performance vastly. We analyze the transferability of point cloud features between two different LiDAR sensor set-ups (32 and 64 vertical scanning planes with different geometry). We propose a supervised training methodology to learn transferable features in a pre-training step on LiDAR datasets that are heterogeneous in their data and label domains. In extensive experiments on object detection and semantic segmentation in a multi-task setup we analyze the performance of our network architecture under the impact of a change in the input data domain. We show that our pre-training approach effectively increases performance for both target tasks at once without having an actual multi-task dataset available for pre-training. [paper]
Pilanet:Analyzing the Cross-Sensor Portability of Neural Network Architectures for LiDAR-based Semantic Labeling
State-of-the-art approaches for the semantic labeling of LiDAR point clouds heavily rely on the use of deep Convolutional Neural Networks (CNNs). However, transferring network architectures across different LiDAR sensor types represents a significant challenge, especially due to sensor specific design choices with regard to network architecture as well as data representation. In this paper we propose a new CNN architecture for the point-wise semantic labeling of LiDAR data which achieves state-of-the-art results while increasing portability across sensor types. This represents a significant advantage given the fast-paced development of LiDAR hardware technology. We perform a thorough quantitative cross-sensor analysis of semantic labeling performance in comparison to a state-of-the-art reference method. Our evaluation shows that the proposed architecture is indeed highly portable, yielding an improvement of 10 percentage points in the Intersectionover-Union (IoU) score when compared to the reference approach. Further, the results indicate that the proposed network architecture can provide an efficient way for the automated generation of large-scale training data for novel LiDAR sensor types without the need for extensive manual annotation or multi-modal label transfer. [paper]
Specialized on-chip accelerators are widely used to improve the energy efficiency of computing systems. Recent advances in memory technology have enabled near-data accelerators (NDAs), which reside off-chip close to main memory and can yield further benefits than on-chip accelerators. However, enforcing coherence with the rest of the system, which is already a major challenge for accelerators, becomes more difficult for NDAs. This is because (1) the cost of communication between NDAs and CPUs is high, and (2) NDA applications generate a lot of off-chip data movement. As a result, as we show in this work, existing coherence mechanisms eliminate most of the benefits of NDAs. We extensively analyze these mechanisms, and observe that (1) the majority of off-chip coherence traffic is unnecessary, and (2) much of the off-chip traffic can be eliminated if a coherence mechanism has insight into the memory accesses performed by the NDA. Based on our observations, we propose CoNDA, a coherence mechanism that lets an NDA optimistically execute an NDA kernel, under the assumption that the NDA has all necessary coherence permissions. This optimistic execution allows CoNDA to gather information on the memory accesses performed by the NDA and by the rest of the system. CoNDA exploits this information to avoid performing unnecessary coherence requests, and thus, significantly reduces data movement for coherence. We evaluate CoNDA using state-of-the-art graph processing and hybrid in-memory database workloads. Averaged across all of our workloads operating on modest data set sizes, CoNDA improves performance by 19.6% over the highest-performance prior coherence mechanism (66.0%/51.7% over a CPU-only/NDA-only system) and reduces memory system energy consumption by 18.0% over the most energy-efficient prior coherence mechanism (43.7% over CPU-only). CoNDA comes within 10.4% and 4.4% of the performance and energy of an ideal mechanism with no cost for coherence. The benefits of CoNDA increase with large data sets, as CoNDA improves performance over the highest-performance prior coherence mechanism by 38.3% (8.4x/7.7x over CPU-only/NDA-only), and comes within 10.2% of an ideal no-cost coherence mechanism. [paper]
Training deep networks with limited labeled data while achieving a strong generalization ability is key in the quest to reduce human annotation efforts. This is the goal of semi-supervised learning, which exploits more widely available unlabeled data to complement small labeled data sets. In this paper, we propose a novel framework for discriminative pixel-level tasks using a generative model of both images and labels. Concretely, we learn a generative adversarial network that captures the joint image-label distribution and is trained efficiently using a large set of unlabeled images supplemented with only few labeled ones. We build our architecture on top of StyleGAN2, augmented with a label synthesis branch. Image labeling at test time is achieved by first embedding the target image into the joint latent space via an encoder network and test-time optimization, and then generating the label from the inferred embedding. We evaluate our approach in two important domains: medical image segmentation and part-based face segmentation. We demonstrate strong in-domain performance compared to several baselines, and are the first to showcase extreme out-of-domain generalization, such as transferring from CT to MRI in medical imaging, and photographs of real faces to paintings, sculptures, and even cartoons and animal faces. [paper]
Recent reports suggest that a generic supervised deep CNN model trained on a large-scale dataset reduces, but does not remove, dataset bias on a standard benchmark. Fine-tuning deep models in a new domain can require a significant amount of data, which for many applications is simply not available. We propose a new CNN architecture which introduces an adaptation layer and an additional domain confusion loss, to learn a representation that is both semantically meaningful and domain invariant. We additionally show that a domain confusion metric can be used for model selection to determine the dimension of an adaptation layer and the best position for the layer in the CNN architecture. Our proposed adaptation method offers empirical performance which exceeds previously published results on a standard benchmark visual domain adaptation task. [paper]
We present a boundary-aware domain adaptation model for LiDAR scan full-scene semantic segmentation (LiDARNet). Our model can extract both the domain private features and the domain shared features with a two branch structure. We embedded Gated-SCNN into the segmentor component of LiDARNet to learn boundary information while learning to predict full-scene semantic segmentation labels. Moreover, we further reduce the domain gap by inducing the model to learn a mapping between two domains using the domain shared and private features. Besides, we introduce a new dataset (SemanticUSL). The dataset has the same data format and ontology as SemanticKITTI. We conducted experiments on real-world datasets SemanticKITTI, SemanticPOSS, and SemanticUSL, which have differences in channel distributions, reflectivity distributions, diversity of scenes, and sensors setup. Using our approach, we can get a single projection-based LiDAR full-scene semantic segmentation model working on both domains. Our model can keep almost the same performance on the source domain after adaptation and get an 8%-22% mIoU performance increase in the target domain. [paper][code] come soon
SqueezeSegV2: Improved Model Structure and Unsupervised Domain Adaptation for Road-Object Segmentation from a LiDAR Point Cloud
Earlier work demonstrates the promise of deep-learning-based approaches for point cloud segmentation; however, these approaches need to be improved to be practically useful. To this end, we introduce a new model SqueezeSegV2. With an improved model structure, SqueezeSetV2 is more robust against dropout noises in LiDAR point cloud and therefore achieves significant accuracy improvement. Training models for point cloud segmentation requires large amounts of labeled data, which is expensive to obtain. To sidestep the cost of data collection and annotation, simulators such as GTA-V can be used to create unlimited amounts of labeled, synthetic data. However, due to domain shift, models trained on synthetic data often do not generalize well to the real world. Existing domain-adaptation methods mainly focus on images and most of them cannot be directly applied to point clouds. We address this problem with a domain-adaptation training pipeline consisting of three major components: 1) learned intensity rendering, 2) geodesic correlation alignment, and 3) progressive domain calibration. When trained on real data, our new model exhibits segmentation accuracy improvements of 6.0-8.6% over the original SqueezeSeg. When training our new model on synthetic data using the proposed domain adaptation pipeline, we nearly double test accuracy on real-world data, from 29.0% to 57.4%. Our source code and synthetic dataset are open sourced. [paper][code]
Domain Adaptation (DA) approaches achieved significant improvements in a wide range of machine learning and computer vision tasks (i.e., classification, detection, and segmentation). However, as far as we are aware, there are few methods yet to achieve domain adaptation directly on 3D point cloud data. The unique challenge of point cloud data lies in its abundant spatial geometric information, and the semantics of the whole object is contributed by including regional geometric structures. Specifically, most general-purpose DA methods that struggle for global feature alignment and ignore local geometric information are not suitable for 3D domain alignment. In this paper, we propose a novel 3D Domain Adaptation Network for point cloud data (PointDAN). PointDAN jointly aligns the global and local features in multi-level. For local alignment, we propose Self-Adaptive (SA) node module with an adjusted receptive field to model the discriminative local structures for aligning domains. To represent hierarchically scaled features, node-attention module is further introduced to weight the relationship of SA nodes across objects and domains. For global alignment, an adversarial-training strategy is employed to learn and align global features across domains. Since there is no common evaluation benchmark for 3D point cloud DA scenario, we build a general benchmark (i.e., PointDA-10) extracted from three popular 3D object/scene datasets (i.e., ModelNet, ShapeNet and ScanNet) for cross-domain 3D objects classification fashion. Extensive experiments on PointDA-10 illustrate the superiority of our model over the state-of-the-art general-purpose DA methods. [paper][code]
We study an unsupervised domain adaptation problem for the semantic labeling of 3D point clouds, with a particular focus on domain discrepancies induced by different LiDAR sensors. Based on the observation that sparse 3D point clouds are sampled from 3D surfaces, we take a Complete and Label approach to recover the underlying surfaces before passing them to a segmentation network. Specifically, we design a Sparse Voxel Completion Network (SVCN) to complete the 3D surfaces of a sparse point cloud. Unlike semantic labels, to obtain training pairs for SVCN requires no manual labeling. We also introduce local adversarial learning to model the surface prior. The recovered 3D surfaces serve as a canonical domain, from which semantic labels can transfer across different LiDAR sensors. Experiments and ablation studies with our new benchmark for cross-domain semantic labeling of LiDAR data show that the proposed approach provides 6.3-37.6% better performance than previous domain adaptation methods. [paper]
Point cloud semantic segmentation is a crucial task in 3D scene understanding. Existing methods mainly focus on employing a large number of annotated labels for supervised semantic segmentation. Nonetheless, manually labeling such large point clouds for the supervised segmentation task is time-consuming. In order to reduce the number of annotated labels, we propose a semi-supervised semantic point cloud segmentation network, named SSPC-Net, where we train the semantic segmentation network by inferring the labels of unlabeled points from the few annotated 3D points. In our method, we first partition the whole point cloud into superpoints and build superpoint graphs to mine the long-range dependencies in point clouds. Based on the constructed superpoint graph, we then develop a dynamic label propagation method to generate the pseudo labels for the unsupervised superpoints. Particularly, we adopt a superpoint dropout strategy to dynamically select the generated pseudo labels. In order to fully exploit the generated pseudo labels of the unsupervised superpoints, we furthermore propose a coupled attention mechanism for superpoint feature embedding. Finally, we employ the cross-entropy loss to train the semantic segmentation network with the labels of the supervised superpoints and the pseudo labels of the unsupervised superpoints. Experiments on various datasets demonstrate that our semi-supervised segmentation method can achieve better performance than the current semi-supervised segmentation method with fewer annotated 3D points. [paper][code]
Unsupervised Domain Adaptation (UDA) is crucial to tackle the lack of annotations in a new domain. There are many multi-modal datasets, but most UDA approaches are uni-modal. In this work, we explore how to learn from multi-modality and propose cross-modal UDA (xMUDA) where we assume the presence of 2D images and 3D point clouds for 3D semantic segmentation. This is challenging as the two input spaces are heterogeneous and can be impacted differently by domain shift. In xMUDA, modalities learn from each other through mutual mimicking, disentangled from the segmentation objective, to prevent the stronger modality from adopting false predictions from the weaker one. We evaluate on new UDA scenarios including day-to-night, country-to-country and dataset-to-dataset, leveraging recent autonomous driving datasets. xMUDA brings large improvements over uni-modal UDA on all tested scenarios, and is complementary to state-of-the-art UDA techniques. [paper][code]
Three-dimensional object detection utilizing LiDAR point cloud data is an indispensable part of autonomous driving perception systems. Point cloud-based 3D object detection has been a better replacement for higher accuracy than cameras during nighttime. However, most LiDAR-based 3D object methods work in a supervised manner, which means their state-of-the-art performance relies heavily on a large-scale and well-labeled dataset, while these annotated datasets could be expensive to obtain and only accessible in the limited scenario. Transfer learning is a promising approach to reduce the large-scale training datasets requirement, but existing transfer learning object detectors are primarily for 2D object detection rather than 3D. In this work, we utilize the 3D point cloud data more effectively by representing the birds-eye-view (BEV) scene and propose a transfer learning based point cloud semantic segmentation for 3D object detection. The proposed model minimizes the need for large-scale training datasets and consequently reduces the training time. First, a preprocessing stage filters the raw point cloud data to a BEV map within a specific field of view. Second, the transfer learning stage uses knowledge from the previously learned classification task (with more data for training) and generalizes the semantic segmentation-based 2D object detection task. Finally, 2D detection results from the BEV image have been back-projected into 3D in the postprocessing stage. We verify results on two datasets: the KITTI 3D object detection dataset and the Ouster LiDAR-64 dataset, thus demonstrating that the proposed method is highly competitive in terms of mean average precision (mAP up to 70%) while still running at more than 30 frames per second (FPS). [paper]
This paper presents a novel CNN-based approach for synthesizing high-resolution LiDAR point cloud data. Our approach generates semantically and perceptually realistic results with guidance from specialized loss-functions. First, we utilize a modified per-point loss that addresses missing LiDAR point measurements. Second, we align the quality of our generated output with real-world sensor data by applying a perceptual loss. In large-scale experiments on real-world datasets, we evaluate both the geometric accuracy and semantic segmentation performance using our generated data vs. ground truth. In a mean opinion score testing we further assess the perceptual quality of our generated point clouds. Our results demonstrate a significant quantitative and qualitative improvement in both geometry and semantics over traditional non CNN-based up-sampling methods. [paper][code]
Building models capable of generating structured output is a key challenge for AI and robotics. While generative models have been explored on many types of data, little work has been done on synthesizing lidar scans, which play a key role in robot mapping and localization. In this work, we show that one can adapt deep generative models for this task by unravelling lidar scans into a 2D point map. Our approach can generate high quality samples, while simultaneously learning a meaningful latent representation of the data. We demonstrate significant improvements against state-of-the-art point cloud generation methods. Furthermore, we propose a novel data representation that augments the 2D signal with absolute positional information. We show that this helps robustness to noisy and imputed input; the learned model can recover the underlying lidar scan from seemingly uninformative data [paper]code]
We propose a noise-aware GAN for generative modeling of 3D LiDAR data on a projected 2D representation (aka spherical projection). Although the 2D representation has been adopted in many LiDAR processing tasks, generative modeling is non-trivial due to the discrete dropout noises caused by LiDAR’s lossy measurement. Our GAN can effectively learn the LiDAR data by representing such discrete data distribution as a composite of two modalities: an underlying complete depth and the corresponding reflective uncertainty. [paper][code]
Inferring semantic information towards an understanding of the surrounding environment is crucial for autonomous vehicles to drive safely. Deep learning-based segmentation methods can infer semantic information directly from laser range data, even in the absence of other sensor modalities such as cameras. In this paper, we address improving the generalization capabilities of such deep learning models to range data that was captured using a different sensor and in situations where no labeled data is available for the new sensor setup. Our approach assists the domain transfer of a LiDARonly semantic segmentation model to a different sensor and environment exploiting existing geometric mapping systems. To this end, we fuse sequential scans in the source dataset into a dense mesh and render semi-synthetic scans that match those of the target sensor setup. Unlike simulation, this approach provides a real-to-real transfer of geometric information and delivers additionally more accurate remission information. We implemented and thoroughly tested our approach by transferring semantic scans between two different real-world datasets with different sensor setups. Our experiments show that we can improve the segmentation performance substantially with zero manual re-labeling. This approach solves the number one feature request since we released our semantic segmentation library LiDAR-bonnetal [paper][code]
LiDAR semantic segmentation provides 3D semantic information about the environment, an essential cue for intelligent systems, such as autonomous vehicles, during their decision making processes. Unfortunately, the annotation process for this task is very expensive. To overcome this, it is key to find models that generalize well or adapt to additional domains where labeled data is limited. This work addresses the problem of unsupervised domain adaptation for LiDAR semantic segmentation models. We propose simple but effective strategies to reduce the domain shift by aligning the data distribution on the input space. Besides, we present a learningbased module to align the distribution of the semantic classes of the target domain to the source domain. Our approach achieves new state-of-the-art results on three different public datasets, which showcase adaptation to three different domains. [paper]
In autonomous driving, a LiDAR-based object detector should perform reliably at different geographic locations and under various weather conditions. While recent 3D detection research focuses on improving performance within a single domain, our study reveals that the performance of modern detectors can drop drastically cross-domain. In this paper, we investigate unsupervised domain adaptation (UDA) for LiDAR-based 3D object detection. On the Waymo Domain Adaptation dataset, we identify the deteriorating point cloud quality as the root cause of the performance drop. To address this issue, we present Semantic Point Generation (SPG), a general approach to enhance the reliability of LiDAR detectors against domain shifts. Specifically, SPG generates semantic points at the predicted foreground regions and faithfully recovers missing parts of the foreground objects, which are caused by phenomena such as occlusions, low reflectance or weather interference. By merging the semantic points with the original points, we obtain an augmented point cloud, which can be directly consumed by modern LiDAR-based detectors. To validate the wide applicability of SPG, we experiment with two representative detectors, PointPillars and PV-RCNN. On the UDA task, SPG significantly improves both detectors across all object categories of interest and at all difficulty levels. SPG can also benefit object detection in the original domain. On the Waymo Open Dataset and KITTI, SPG improves 3D detection results of these two methods across all categories. Combined with PV-RCNN, SPG achieves state-of-the-art 3D detection results on KITTI. [paper]
We present a new domain adaptive self-training pipeline, named ST3D, for unsupervised domain adaptation on 3D object detection from point clouds. First, we pre-train the 3D detector on the source domain with our proposed random object scaling strategy for mitigating the negative effects of source domain bias. Then, the detector is iteratively improved on the target domain by alternatively conducting two steps, which are the pseudo label updating with the developed quality-aware triplet memory bank and the model training with curriculum data augmentation. These specific designs for 3D object detection enable the detector to be trained with consistent and high-quality pseudo labels and to avoid overfitting to the large number of easy examples in pseudo labeled data. Our ST3D achieves stateof-the-art performance on all evaluated datasets and even surpasses fully supervised results on KITTI 3D object detection benchmark. [Code][paper]
Semantic segmentation is a fundamental task in understanding urban mobile laser scanning (MLS) point clouds. Recently, deep learning-based methods have become prominent for semantic segmentation of MLS point clouds, and many recent works have achieved state-of-the-art performance on open benchmarks. However, due to differences of objects across different scenes such as different height of buildings and different forms of the same road-side objects, the existing open benchmarks (namely source scenes) are often significantly different from the actual application datasets (namely target scenes). This results in underperformance of semantic segmentation networks trained using source scenes when applied to target scenes. In this paper, we propose a novel method to perform unsupervised scene adaptation for semantic segmentation of urban MLS point clouds. Firstly, we show the scene transfer phenomena in urban MLS point clouds. Then, we propose a new pointwise attentive transformation module (PW-ATM) to adaptively perform the data alignment. Next, a maximum classifier discrepancy-based (MCD-based) adversarial learning framework is adopted to further achieve feature alignment. Finally, an end-to-end alignment deep network architecture is designed for the unsupervised scene adaptation semantic segmentation of urban MLS point clouds. To experimentally evaluate the performance of our proposed approach, two large-scale labeled source scenes and two different target scenes were used for the training. Moreover, four actual application scenes are used for the testing. The experimental results indicated that our approach can effectively achieve scene adaptation for semantic segmentation of urban MLS point clouds. [paper]
Top-performing deep architectures are trained on massive amounts of labeled data. In the absence of labeled data for a certain task, domain adaptation often provides an attractive option given that labeled data of similar nature but from a different domain (e.g. synthetic images) are available. Here, we propose a new approach to domain adaptation in deep architectures that can be trained on large amount of labeled data from the source domain and large amount of unlabeled data from the target domain (no labeled targetdomain data is necessary). As the training progresses, the approach promotes the emergence of “deep” features that are (i) discriminative for the main learning task on the source domain and (ii) invariant with respect to the shift between the domains. We show that this adaptation behaviour can be achieved in almost any feed-forward model by augmenting it with few standard layers and a simple new gradient reversal layer. The resulting augmented architecture can be trained using standard backpropagation. Overall, the approach can be implemented with little effort using any of the deep-learning packages. The method performs very well in a series of image classification experiments, achieving adaptation effect in the presence of big domain shifts and outperforming previous state-ofthe-art on Office datasets. [paper]
3D LiDAR semantic segmentation is fundamental for autonomous driving. Several Unsupervised Domain Adaptation (UDA) methods for point cloud data have been recently proposed to improve model generalization for different sensors and environments. Researchers working on UDA problems in the image domain have shown that sample mixing can mitigate domain shift. We propose a new approach of sample mixing for point cloud UDA, namely Compositional Semantic Mix (CoSMix), the first UDA approach for point cloud segmentation based on sample mixing. CoSMix consists of a two-branch symmetric network that can process labelled synthetic data (source) and real-world unlabelled point clouds (target) concurrently. Each branch operates on one domain by mixing selected pieces of data from the other one, and by using the semantic information derived from source labels and target pseudo-labels. We evaluate CoSMix on two large-scale datasets, showing that it outperforms state-of-the-art methods by a large margin. codepaper
Paris-Lille-3D: a large and high-quality ground truth urban point cloud dataset for automatic segmentation and classification
This paper introduces a new Urban Point Cloud Dataset for Automatic Segmentation and Classification acquired by Mobile Laser Scanning (MLS). We describe how the dataset is obtained from acquisition to post-processing and labeling. This dataset can be used to learn classification algorithm, however, given that a great attention has been paid to the split between the different objects, this dataset can also be used to learn the segmentation. The dataset consists of around 2km of MLS point cloud acquired in two cities. The number of points and range of classes make us consider that it can be used to train Deep-Learning methods. Besides we show some results of automatic segmentation and classification. [paper][code]