Slobodan Ilic

Zheng Qin, Hao Yu, Changjian Wang et al.

We study the problem of extracting accurate correspondences for point cloud registration. Recent keypoint-free methods have shown great potential through bypassing the detection of repeatable keypoints which is difficult to do especially in low-overlap scenarios. They seek correspondences over downsampled superpoints, which are then propagated to dense points. Superpoints are matched based on whether their neighboring patches overlap. Such sparse and loose matching requires contextual features capturing the geometric structure of the point clouds. We propose Geometric Transformer, or GeoTransformer for short, to learn geometric feature for robust superpoint matching. It encodes pair-wise distances and triplet-wise angles, making it invariant to rigid transformation and robust in low-overlap cases. The simplistic design attains surprisingly high matching accuracy such that no RANSAC is required in the estimation of alignment transformation, leading to $100$ times acceleration. Extensive experiments on rich benchmarks encompassing indoor, outdoor, synthetic, multiway and non-rigid demonstrate the efficacy of GeoTransformer. Notably, our method improves the inlier ratio by $18{\sim}31$ percentage points and the registration recall by over $7$ points on the challenging 3DLoMatch benchmark. Our code and models are available at \url{https://github.com/qinzheng93/GeoTransformer}.

HyunJun Jung, Patrick Ruhkamp, Guangyao Zhai et al.

Learning-based methods to solve dense 3D vision problems typically train on 3D sensor data. The respectively used principle of measuring distances provides advantages and drawbacks. These are typically not compared nor discussed in the literature due to a lack of multi-modal datasets. Texture-less regions are problematic for structure from motion and stereo, reflective material poses issues for active sensing, and distances for translucent objects are intricate to measure with existing hardware. Training on inaccurate or corrupt data induces model bias and hampers generalisation capabilities. These effects remain unnoticed if the sensor measurement is considered as ground truth during the evaluation. This paper investigates the effect of sensor errors for the dense 3D vision tasks of depth estimation and reconstruction. We rigorously show the significant impact of sensor characteristics on the learned predictions and notice generalisation issues arising from various technologies in everyday household environments. For evaluation, we introduce a carefully designed dataset\footnote{dataset available at https://github.com/Junggy/HAMMER-dataset} comprising measurements from commodity sensors, namely D-ToF, I-ToF, passive/active stereo, and monocular RGB+P. Our study quantifies the considerable sensor noise impact and paves the way to improved dense vision estimates and targeted data fusion.

Agnieszka Tomczak, Aarushi Gupta, Slobodan Ilic et al. · eth-zurich

Generative adversarial networks (GANs) offer an effective solution to the image-to-image translation problem, thereby allowing for new possibilities in medical imaging. They can translate images from one imaging modality to another at a low cost. For unpaired datasets, they rely mostly on cycle loss. Despite its effectiveness in learning the underlying data distribution, it can lead to a discrepancy between input and output data. The purpose of this work is to investigate the hypothesis that we can predict image quality based on its latent representation in the GANs bottleneck. We achieve this by corrupting the latent representation with noise and generating multiple outputs. The degree of differences between them is interpreted as the strength of the representation: the more robust the latent representation, the fewer changes in the output image the corruption causes. Our results demonstrate that our proposed method has the ability to i) predict uncertain parts of synthesized images, and ii) identify samples that may not be reliable for downstream tasks, e.g., liver segmentation task.

Hao Yu, Zheng Qin, Ji Hou et al.

The intrinsic rotation invariance lies at the core of matching point clouds with handcrafted descriptors. However, it is widely despised by recent deep matchers that obtain the rotation invariance extrinsically via data augmentation. As the finite number of augmented rotations can never span the continuous SO(3) space, these methods usually show instability when facing rotations that are rarely seen. To this end, we introduce RoITr, a Rotation-Invariant Transformer to cope with the pose variations in the point cloud matching task. We contribute both on the local and global levels. Starting from the local level, we introduce an attention mechanism embedded with Point Pair Feature (PPF)-based coordinates to describe the pose-invariant geometry, upon which a novel attention-based encoder-decoder architecture is constructed. We further propose a global transformer with rotation-invariant cross-frame spatial awareness learned by the self-attention mechanism, which significantly improves the feature distinctiveness and makes the model robust with respect to the low overlap. Experiments are conducted on both the rigid and non-rigid public benchmarks, where RoITr outperforms all the state-of-the-art models by a considerable margin in the low-overlapping scenarios. Especially when the rotations are enlarged on the challenging 3DLoMatch benchmark, RoITr surpasses the existing methods by at least 13 and 5 percentage points in terms of Inlier Ratio and Registration Recall, respectively.

Ivan Shugurov, Fu Li, Benjamin Busam et al.

We present a novel one-shot method for object detection and 6 DoF pose estimation, that does not require training on target objects. At test time, it takes as input a target image and a textured 3D query model. The core idea is to represent a 3D model with a number of 2D templates rendered from different viewpoints. This enables CNN-based direct dense feature extraction and matching. The object is first localized in 2D, then its approximate viewpoint is estimated, followed by dense 2D-3D correspondence prediction. The final pose is computed with PnP. We evaluate the method on LineMOD, Occlusion, Homebrewed, YCB-V and TLESS datasets and report very competitive performance in comparison to the state-of-the-art methods trained on synthetic data, even though our method is not trained on the object models used for testing.

Fu Li, Hao Yu, Ivan Shugurov et al.

Pose estimation of 3D objects in monocular images is a fundamental and long-standing problem in computer vision. Existing deep learning approaches for 6D pose estimation typically rely on the assumption of availability of 3D object models and 6D pose annotations. However, precise annotation of 6D poses in real data is intricate, time-consuming and not scalable, while synthetic data scales well but lacks realism. To avoid these problems, we present a weakly-supervised reconstruction-based pipeline, named NeRF-Pose, which needs only 2D object segmentation and known relative camera poses during training. Following the first-reconstruct-then-regress idea, we first reconstruct the objects from multiple views in the form of an implicit neural representation. Then, we train a pose regression network to predict pixel-wise 2D-3D correspondences between images and the reconstructed model. At inference, the approach only needs a single image as input. A NeRF-enabled PnP+RANSAC algorithm is used to estimate stable and accurate pose from the predicted correspondences. Experiments on LineMod and LineMod-Occlusion show that the proposed method has state-of-the-art accuracy in comparison to the best 6D pose estimation methods in spite of being trained only with weak labels. Besides, we extend the Homebrewed DB dataset with more real training images to support the weakly supervised task and achieve compelling results on this dataset. The extended dataset and code will be released soon.

Ivan Shugurov, Sergey Zakharov, Slobodan Ilic

We propose a three-stage 6 DoF object detection method called DPODv2 (Dense Pose Object Detector) that relies on dense correspondences. We combine a 2D object detector with a dense correspondence estimation network and a multi-view pose refinement method to estimate a full 6 DoF pose. Unlike other deep learning methods that are typically restricted to monocular RGB images, we propose a unified deep learning network allowing different imaging modalities to be used (RGB or Depth). Moreover, we propose a novel pose refinement method, that is based on differentiable rendering. The main concept is to compare predicted and rendered correspondences in multiple views to obtain a pose which is consistent with predicted correspondences in all views. Our proposed method is evaluated rigorously on different data modalities and types of training data in a controlled setup. The main conclusions is that RGB excels in correspondence estimation, while depth contributes to the pose accuracy if good 3D-3D correspondences are available. Naturally, their combination achieves the overall best performance. We perform an extensive evaluation and an ablation study to analyze and validate the results on several challenging datasets. DPODv2 achieves excellent results on all of them while still remaining fast and scalable independent of the used data modality and the type of training data

Hao Yu, Ji Hou, Zheng Qin et al.

Successful point cloud registration relies on accurate correspondences established upon powerful descriptors. However, existing neural descriptors either leverage a rotation-variant backbone whose performance declines under large rotations, or encode local geometry that is less distinctive. To address this issue, we introduce RIGA to learn descriptors that are Rotation-Invariant by design and Globally-Aware. From the Point Pair Features (PPFs) of sparse local regions, rotation-invariant local geometry is encoded into geometric descriptors. Global awareness of 3D structures and geometric context is subsequently incorporated, both in a rotation-invariant fashion. More specifically, 3D structures of the whole frame are first represented by our global PPF signatures, from which structural descriptors are learned to help geometric descriptors sense the 3D world beyond local regions. Geometric context from the whole scene is then globally aggregated into descriptors. Finally, the description of sparse regions is interpolated to dense point descriptors, from which correspondences are extracted for registration. To validate our approach, we conduct extensive experiments on both object- and scene-level data. With large rotations, RIGA surpasses the state-of-the-art methods by a margin of 8\degree in terms of the Relative Rotation Error on ModelNet40 and improves the Feature Matching Recall by at least 5 percentage points on 3DLoMatch.

HyunJun Jung, Patrick Ruhkamp, Guangyao Zhai et al.

Depth estimation is a core task in 3D computer vision. Recent methods investigate the task of monocular depth trained with various depth sensor modalities. Every sensor has its advantages and drawbacks caused by the nature of estimates. In the literature, mostly mean average error of the depth is investigated and sensor capabilities are typically not discussed. Especially indoor environments, however, pose challenges for some devices. Textureless regions pose challenges for structure from motion, reflective materials are problematic for active sensing, and distances for translucent material are intricate to measure with existing sensors. This paper proposes HAMMER, a dataset comprising depth estimates from multiple commonly used sensors for indoor depth estimation, namely ToF, stereo, structured light together with monocular RGB+P data. We construct highly reliable ground truth depth maps with the help of 3D scanners and aligned renderings. A popular depth estimators is trained on this data and typical depth senosors. The estimates are extensively analyze on different scene structures. We notice generalization issues arising from various sensor technologies in household environments with challenging but everyday scene content. HAMMER, which we make publicly available, provides a reliable base to pave the way to targeted depth improvements and sensor fusion approaches.

Ivan Shugurov, Ivan Pavlov, Sergey Zakharov et al.

This paper introduces a novel multi-view 6 DoF object pose refinement approach focusing on improving methods trained on synthetic data. It is based on the DPOD detector, which produces dense 2D-3D correspondences between the model vertices and the image pixels in each frame. We have opted for the use of multiple frames with known relative camera transformations, as it allows introduction of geometrical constraints via an interpretable ICP-like loss function. The loss function is implemented with a differentiable renderer and is optimized iteratively. We also demonstrate that a full detection and refinement pipeline, which is trained solely on synthetic data, can be used for auto-labeling real data. We perform quantitative evaluation on LineMOD, Occlusion, Homebrewed and YCB-V datasets and report excellent performance in comparison to the state-of-the-art methods trained on the synthetic and real data. We demonstrate empirically that our approach requires only a few frames and is robust to close camera locations and noise in extrinsic camera calibration, making its practical usage easier and more ubiquitous.

Alexander Baumann, Leonardo Ayala, Alexander Studier-Fischer et al.

Hyperspectral imaging (HSI) is emerging as a promising novel imaging modality with various potential surgical applications. Currently available cameras, however, suffer from poor integration into the clinical workflow because they require the lights to be switched off, or the camera to be manually recalibrated as soon as lighting conditions change. Given this critical bottleneck, the contribution of this paper is threefold: (1) We demonstrate that dynamically changing lighting conditions in the operating room dramatically affect the performance of HSI applications, namely physiological parameter estimation, and surgical scene segmentation. (2) We propose a novel learning-based approach to automatically recalibrating hyperspectral images during surgery and show that it is sufficiently accurate to replace the tedious process of white reference-based recalibration. (3) Based on a total of 742 HSI cubes from a phantom, porcine models, and rats we show that our recalibration method not only outperforms previously proposed methods, but also generalizes across species, lighting conditions, and image processing tasks. Due to its simple workflow integration as well as high accuracy, speed, and generalization capabilities, our method could evolve as a central component in clinical surgical HSI.

Junwen Huang, Hao Yu, Kuan-Ting Yu et al.

Recent learning methods for object pose estimation require resource-intensive training for each individual object instance or category, hampering their scalability in real applications when confronted with previously unseen objects. In this paper, we propose MatchU, a Fuse-Describe-Match strategy for 6D pose estimation from RGB-D images. MatchU is a generic approach that fuses 2D texture and 3D geometric cues for 6D pose prediction of unseen objects. We rely on learning geometric 3D descriptors that are rotation-invariant by design. By encoding pose-agnostic geometry, the learned descriptors naturally generalize to unseen objects and capture symmetries. To tackle ambiguous associations using 3D geometry only, we fuse additional RGB information into our descriptor. This is achieved through a novel attention-based mechanism that fuses cross-modal information, together with a matching loss that leverages the latent space learned from RGB data to guide the descriptor learning process. Extensive experiments reveal the generalizability of both the RGB-D fusion strategy as well as the descriptor efficacy. Benefiting from the novel designs, MatchU surpasses all existing methods by a significant margin in terms of both accuracy and speed, even without the requirement of expensive re-training or rendering.

Dekai Zhu, Yixuan Hu, Youquan Liu et al.

Leveraging recent diffusion models, LiDAR-based large-scale 3D scene generation has achieved great success. While recent voxel-based approaches can generate both geometric structures and semantic labels, existing range-view methods are limited to producing unlabeled LiDAR scenes. Relying on pretrained segmentation models to predict the semantic maps often results in suboptimal cross-modal consistency. To address this limitation while preserving the advantages of range-view representations, such as computational efficiency and simplified network design, we propose Spiral, a novel range-view LiDAR diffusion model that simultaneously generates depth, reflectance images, and semantic maps. Furthermore, we introduce novel semantic-aware metrics to evaluate the quality of the generated labeled range-view data. Experiments on the SemanticKITTI and nuScenes datasets demonstrate that Spiral achieves state-of-the-art performance with the smallest parameter size, outperforming two-step methods that combine the generative and segmentation models. Additionally, we validate that range images generated by Spiral can be effectively used for synthetic data augmentation in the downstream segmentation training, significantly reducing the labeling effort on LiDAR data.

Dekai Zhu, Yan Di, Stefan Gavranovic et al.

Denoising diffusion probabilistic models have achieved significant success in point cloud generation, enabling numerous downstream applications, such as generative data augmentation and 3D model editing. However, little attention has been given to generating point clouds with point-wise segmentation labels, as well as to developing evaluation metrics for this task. Therefore, in this paper, we present SeaLion, a novel diffusion model designed to generate high-quality and diverse point clouds with fine-grained segmentation labels. Specifically, we introduce the semantic part-aware latent point diffusion technique, which leverages the intermediate features of the generative models to jointly predict the noise for perturbed latent points and associated part segmentation labels during the denoising process, and subsequently decodes the latent points to point clouds conditioned on part segmentation labels. To effectively evaluate the quality of generated point clouds, we introduce a novel point cloud pairwise distance calculation method named part-aware Chamfer distance (p-CD). This method enables existing metrics, such as 1-NNA, to measure both the local structural quality and inter-part coherence of generated point clouds. Experiments on the large-scale synthetic dataset ShapeNet and real-world medical dataset IntrA demonstrate that SeaLion achieves remarkable performance in generation quality and diversity, outperforming the existing state-of-the-art model, DiffFacto, by 13.33% and 6.52% on 1-NNA (p-CD) across the two datasets. Experimental analysis shows that SeaLion can be trained semi-supervised, thereby reducing the demand for labeling efforts. Lastly, we validate the applicability of SeaLion in generative data augmentation for training segmentation models and the capability of SeaLion to serve as a tool for part-aware 3D shape editing.

Dekai Zhu, Stefan Gavranovic, Flavien Boussuge et al.

Data augmentation is widely used to train deep learning models to address data scarcity. However, traditional data augmentation (TDA) typically relies on simple geometric transformation, such as random rotation and rescaling, resulting in minimal data diversity enrichment and limited model performance improvement. State-of-the-art generative models for 3D shape generation rely on the denoising diffusion probabilistic models and manage to generate realistic novel point clouds for 3D content creation and manipulation. Nevertheless, the generated 3D shapes lack associated point-wise semantic labels, restricting their usage in enlarging the training data for point cloud segmentation tasks. To bridge the gap between data augmentation techniques and the advanced diffusion models, we extend the state-of-the-art 3D diffusion model, Lion, to a part-aware generative model that can generate high-quality point clouds conditioned on given segmentation masks. Leveraging the novel generative model, we introduce a 3-step generative data augmentation (GDA) pipeline for point cloud segmentation training. Our GDA approach requires only a small amount of labeled samples but enriches the training data with generated variants and pseudo-labeled samples, which are validated by a novel diffusion-based pseudo-label filtering method. Extensive experiments on two large-scale synthetic datasets and a real-world medical dataset demonstrate that our GDA method outperforms TDA approach and related semi-supervised and self-supervised methods.

Jiaqi Hu, Hongli Xu, Junwen Huang et al.

Accurate 6D pose estimation is essential for robotic manipulation in industrial environments. Existing pipelines typically rely on off-the-shelf object detectors followed by cropping and pose refinement, but their performance degrades under challenging conditions such as clutter, poor lighting, and complex backgrounds, making detection the critical bottleneck. In this work, we introduce a standardized and plug-in pipeline for 2D detection of unseen objects in industrial settings. Based on current SOTA baselines, our approach reduces domain shift and background artifacts through low-light image enhancement and background removal guided by open-vocabulary detection with foundation models. This design suppresses the false positives prevalent in raw SAM outputs, yielding more reliable detections for downstream pose estimation. Extensive experiments on real-world industrial bin-picking benchmarks from BOP demonstrate that our method significantly boosts detection accuracy while incurring negligible inference overhead, showing the effectiveness and practicality of the proposed method.

Junwen Huang, Shishir Reddy Vutukur, Peter KT Yu et al.

Typical template-based object pose pipelines estimate the pose by retrieving the closest matching template and aligning it with the observed image. However, failure to retrieve the correct template often leads to inaccurate pose predictions. To address this, we reformulate template-based object pose estimation as a ray alignment problem, where the viewing directions from multiple posed template images are learned to align with a non-posed query image. Inspired by recent progress in diffusion-based camera pose estimation, we embed this formulation into a diffusion transformer architecture that aligns a query image with a set of posed templates. We reparameterize object rotation using object-centered camera rays and model object translation by extending scale-invariant translation estimation to dense translation offsets. Our model leverages geometric priors from the templates to guide accurate query pose inference. A coarse-to-fine training strategy based on narrowed template sampling improves performance without modifying the network architecture. Extensive experiments across multiple benchmark datasets show competitive results of our method compared to state-of-the-art approaches in unseen object pose estimation.

Alexander Baumann, Leonardo Ayala, Silvia Seidlitz et al.

Spectral imaging offers promising applications across diverse domains, including medicine and urban scene understanding, and is already established as a critical modality in remote sensing. However, variability in channel dimensionality and captured wavelengths among spectral cameras impede the development of AI-driven methodologies, leading to camera-specific models with limited generalizability and inadequate cross-camera applicability. To address this bottleneck, we introduce CARL, a model for Camera-Agnostic Representation Learning across RGB, multispectral, and hyperspectral imaging modalities. To enable the conversion of a spectral image with any channel dimensionality to a camera-agnostic representation, we introduce a novel spectral encoder, featuring a self-attention-cross-attention mechanism, to distill salient spectral information into learned spectral representations. Spatio-spectral pre-training is achieved with a novel feature-based self-supervision strategy tailored to CARL. Large-scale experiments across the domains of medical imaging, autonomous driving, and satellite imaging demonstrate our model's unique robustness to spectral heterogeneity, outperforming on datasets with simulated and real-world cross-camera spectral variations. The scalability and versatility of the proposed approach position our model as a backbone for future spectral foundation models.

Shishir Reddy Vutukur, Heike Brock, Benjamin Busam et al.

Object Pose Estimation is a crucial component in robotic grasping and augmented reality. Learning based approaches typically require training data from a highly accurate CAD model or labeled training data acquired using a complex setup. We address this by learning to estimate pose from weakly labeled data without a known CAD model. We propose to use a NeRF to learn object shape implicitly which is later used to learn view-invariant features in conjunction with CNN using a contrastive loss. While NeRF helps in learning features that are view-consistent, CNN ensures that the learned features respect symmetry. During inference, CNN is used to predict view-invariant features which can be used to establish correspondences with the implicit 3d model in NeRF. The correspondences are then used to estimate the pose in the reference frame of NeRF. Our approach can also handle symmetric objects unlike other approaches using a similar training setup. Specifically, we learn viewpoint invariant, discriminative features using NeRF which are later used for pose estimation. We evaluated our approach on LM, LM-Occlusion, and T-Less dataset and achieved benchmark accuracy despite using weakly labeled data.

Hao Yu, Fu Li, Mahdi Saleh et al.

We study the problem of extracting correspondences between a pair of point clouds for registration. For correspondence retrieval, existing works benefit from matching sparse keypoints detected from dense points but usually struggle to guarantee their repeatability. To address this issue, we present CoFiNet - Coarse-to-Fine Network which extracts hierarchical correspondences from coarse to fine without keypoint detection. On a coarse scale and guided by a weighting scheme, our model firstly learns to match down-sampled nodes whose vicinity points share more overlap, which significantly shrinks the search space of a consecutive stage. On a finer scale, node proposals are consecutively expanded to patches that consist of groups of points together with associated descriptors. Point correspondences are then refined from the overlap areas of corresponding patches, by a density-adaptive matching module capable to deal with varying point density. Extensive evaluation of CoFiNet on both indoor and outdoor standard benchmarks shows our superiority over existing methods. Especially on 3DLoMatch where point clouds share less overlap, CoFiNet significantly outperforms state-of-the-art approaches by at least 5% on Registration Recall, with at most two-third of their parameters.

Yehya Abouelnaga, Mai Bui, Slobodan Ilic

We propose a lightweight retrieval-based pipeline to predict 6DOF camera poses from RGB images. Our pipeline uses a convolutional neural network (CNN) to encode a query image as a feature vector. A nearest neighbor lookup finds the pose-wise nearest database image. A siamese convolutional neural network regresses the relative pose from the nearest neighboring database image to the query image. The relative pose is then applied to the nearest neighboring absolute pose to obtain the query image's final absolute pose prediction. Our model is a distilled version of NN-Net that reduces its parameters by 98.87%, information retrieval feature vector size by 87.5%, and inference time by 89.18% without a significant decrease in localization accuracy.

Haowen Deng, Mai Bui, Nassir Navab et al.

In this work, we introduce Deep Bingham Networks (DBN), a generic framework that can naturally handle pose-related uncertainties and ambiguities arising in almost all real life applications concerning 3D data. While existing works strive to find a single solution to the pose estimation problem, we make peace with the ambiguities causing high uncertainty around which solutions to identify as the best. Instead, we report a family of poses which capture the nature of the solution space. DBN extends the state of the art direct pose regression networks by (i) a multi-hypotheses prediction head which can yield different distribution modes; and (ii) novel loss functions that benefit from Bingham distributions on rotations. This way, DBN can work both in unambiguous cases providing uncertainty information, and in ambiguous scenes where an uncertainty per mode is desired. On a technical front, our network regresses continuous Bingham mixture models and is applicable to both 2D data such as images and to 3D data such as point clouds. We proposed new training strategies so as to avoid mode or posterior collapse during training and to improve numerical stability. Our methods are thoroughly tested on two different applications exploiting two different modalities: (i) 6D camera relocalization from images; and (ii) object pose estimation from 3D point clouds, demonstrating decent advantages over the state of the art. For the former we contributed our own dataset composed of five indoor scenes where it is unavoidable to capture images corresponding to views that are hard to uniquely identify. For the latter we achieve the top results especially for symmetric objects of ModelNet dataset.

Giorgia Pitteri, Aurélie Bugeau, Slobodan Ilic et al.

We present an approach for detecting and estimating the 3D poses of objects in images that requires only an untextured CAD model and no training phase for new objects. Our approach combines Deep Learning and 3D geometry: It relies on an embedding of local 3D geometry to match the CAD models to the input images. For points at the surface of objects, this embedding can be computed directly from the CAD model; for image locations, we learn to predict it from the image itself. This establishes correspondences between 3D points on the CAD model and 2D locations of the input images. However, many of these correspondences are ambiguous as many points may have similar local geometries. We show that we can use Mask-RCNN in a class-agnostic way to detect the new objects without retraining and thus drastically limit the number of possible correspondences. We can then robustly estimate a 3D pose from these discriminative correspondences using a RANSAC- like algorithm. We demonstrate the performance of this approach on the T-LESS dataset, by using a small number of objects to learn the embedding and testing it on the other objects. Our experiments show that our method is on par or better than previous methods.

Mai Bui, Tolga Birdal, Haowen Deng et al.

We present a multimodal camera relocalization framework that captures ambiguities and uncertainties with continuous mixture models defined on the manifold of camera poses. In highly ambiguous environments, which can easily arise due to symmetries and repetitive structures in the scene, computing one plausible solution (what most state-of-the-art methods currently regress) may not be sufficient. Instead we predict multiple camera pose hypotheses as well as the respective uncertainty for each prediction. Towards this aim, we use Bingham distributions, to model the orientation of the camera pose, and a multivariate Gaussian to model the position, with an end-to-end deep neural network. By incorporating a Winner-Takes-All training scheme, we finally obtain a mixture model that is well suited for explaining ambiguities in the scene, yet does not suffer from mode collapse, a common problem with mixture density networks. We introduce a new dataset specifically designed to foster camera localization research in ambiguous environments and exhaustively evaluate our method on synthetic as well as real data on both ambiguous scenes and on non-ambiguous benchmark datasets. We plan to release our code and dataset under $\href{https://multimodal3dvision.github.io}{multimodal3dvision.github.io}$.

Wadim Kehl, Federico Tombari, Slobodan Ilic et al.

We present a novel method to track 3D models in color and depth data. To this end, we introduce approximations that accelerate the state-of-the-art in region-based tracking by an order of magnitude while retaining similar accuracy. Furthermore, we show how the method can be made more robust in the presence of depth data and consequently formulate a new joint contour and ICP tracking energy. We present better results than the state-of-the-art while being much faster then most other methods and achieving all of the above on a single CPU core.

Giorgia Pitteri, Slobodan Ilic, Vincent Lepetit

We present a novel approach to the detection and 3D pose estimation of objects in color images. Its main contribution is that it does not require any training phases nor data for new objects, while state-of-the-art methods typically require hours of training time and hundreds of training registered images. Instead, our method relies only on the objects' geometries. Our method focuses on objects with prominent corners, which covers a large number of industrial objects. We first learn to detect object corners of various shapes in images and also to predict their 3D poses, by using training images of a small set of objects. To detect a new object in a given image, we first identify its corners from its CAD model; we also detect the corners visible in the image and predict their 3D poses. We then introduce a RANSAC-like algorithm that robustly and efficiently detects and estimates the object's 3D pose by matching its corners on the CAD model with their detected counterparts in the image. Because we also estimate the 3D poses of the corners in the image, detecting only 1 or 2 corners is sufficient to estimate the pose of the object, which makes the approach robust to occlusions. We finally rely on a final check that exploits the full 3D geometry of the objects, in case multiple objects have the same corner spatial arrangement. The advantages of our approach make it particularly attractive for industrial contexts, and we demonstrate our approach on the challenging T-LESS dataset.

Giorgia Pitteri, Michaël Ramamonjisoa, Slobodan Ilic et al.

Objects with symmetries are common in our daily life and in industrial contexts, but are often ignored in the recent literature on 6D pose estimation from images. In this paper, we study in an analytical way the link between the symmetries of a 3D object and its appearance in images. We explain why symmetrical objects can be a challenge when training machine learning algorithms that aim at estimating their 6D pose from images. We propose an efficient and simple solution that relies on the normalization of the pose rotation. Our approach is general and can be used with any 6D pose estimation algorithm. Moreover, our method is also beneficial for objects that are 'almost symmetrical', i.e. objects for which only a detail breaks the symmetry. We validate our approach within a Faster-RCNN framework on a synthetic dataset made with objects from the T-Less dataset, which exhibit various types of symmetries, as well as real sequences from T-Less.

Sergey Zakharov, Wadim Kehl, Benjamin Planche et al.

In this paper, we address the problem of 3D object instance recognition and pose estimation of localized objects in cluttered environments using convolutional neural networks. Inspired by the descriptor learning approach of Wohlhart et al., we propose a method that introduces the dynamic margin in the manifold learning triplet loss function. Such a loss function is designed to map images of different objects under different poses to a lower-dimensional, similarity-preserving descriptor space on which efficient nearest neighbor search algorithms can be applied. Introducing the dynamic margin allows for faster training times and better accuracy of the resulting low-dimensional manifolds. Furthermore, we contribute the following: adding in-plane rotations (ignored by the baseline method) to the training, proposing new background noise types that help to better mimic realistic scenarios and improve accuracy with respect to clutter, adding surface normals as another powerful image modality representing an object surface leading to better performance than merely depth, and finally implementing an efficient online batch generation that allows for better variability during the training phase. We perform an exhaustive evaluation to demonstrate the effects of our contributions. Additionally, we assess the performance of the algorithm on the large BigBIRD dataset to demonstrate good scalability properties of the pipeline with respect to the number of models.

Haowen Deng, Tolga Birdal, Slobodan Ilic

We present a novel, data driven approach for solving the problem of registration of two point cloud scans. Our approach is direct in the sense that a single pair of corresponding local patches already provides the necessary transformation cue for the global registration. To achieve that, we first endow the state of the art PPF-FoldNet auto-encoder (AE) with a pose-variant sibling, where the discrepancy between the two leads to pose-specific descriptors. Based upon this, we introduce RelativeNet, a relative pose estimation network to assign correspondence-specific orientations to the keypoints, eliminating any local reference frame computations. Finally, we devise a simple yet effective hypothesize-and-verify algorithm to quickly use the predictions and align two point sets. Our extensive quantitative and qualitative experiments suggests that our approach outperforms the state of the art in challenging real datasets of pairwise registration and that augmenting the keypoints with local pose information leads to better generalization and a dramatic speed-up.

Roman Kaskman, Sergey Zakharov, Ivan Shugurov et al.

Among the most important prerequisites for creating and evaluating 6D object pose detectors are datasets with labeled 6D poses. With the advent of deep learning, demand for such datasets is growing continuously. Despite the fact that some of exist, they are scarce and typically have restricted setups, such as a single object per sequence, or they focus on specific object types, such as textureless industrial parts. Besides, two significant components are often ignored: training using only available 3D models instead of real data and scalability, i.e. training one method to detect all objects rather than training one detector per object. Other challenges, such as occlusions, changing light conditions and changes in object appearance, as well precisely defined benchmarks are either not present or are scattered among different datasets. In this paper we present a dataset for 6D pose estimation that covers the above-mentioned challenges, mainly targeting training from 3D models (both textured and textureless), scalability, occlusions, and changes in light conditions and object appearance. The dataset features 33 objects (17 toy, 8 household and 8 industry-relevant objects) over 13 scenes of various difficulty. We also present a set of benchmarks to test various desired detector properties, particularly focusing on scalability with respect to the number of objects and resistance to changing light conditions, occlusions and clutter. We also set a baseline for the presented benchmarks using a state-of-the-art DPOD detector. Considering the difficulty of making such datasets, we plan to release the code allowing other researchers to extend this dataset or make their own datasets in the future.

Sergey Zakharov, Wadim Kehl, Slobodan Ilic

We present a novel approach to tackle domain adaptation between synthetic and real data. Instead, of employing "blind" domain randomization, i.e., augmenting synthetic renderings with random backgrounds or changing illumination and colorization, we leverage the task network as its own adversarial guide toward useful augmentations that maximize the uncertainty of the output. To this end, we design a min-max optimization scheme where a given task competes against a special deception network to minimize the task error subject to the specific constraints enforced by the deceiver. The deception network samples from a family of differentiable pixel-level perturbations and exploits the task architecture to find the most destructive augmentations. Unlike GAN-based approaches that require unlabeled data from the target domain, our method achieves robust mappings that scale well to multiple target distributions from source data alone. We apply our framework to the tasks of digit recognition on enhanced MNIST variants, classification and object pose estimation on the Cropped LineMOD dataset as well as semantic segmentation on the Cityscapes dataset and compare it to a number of domain adaptation approaches, thereby demonstrating similar results with superior generalization capabilities.

Mai Bui, Christoph Baur, Nassir Navab et al.

Despite recent advances on the topic of direct camera pose regression using neural networks, accurately estimating the camera pose of a single RGB image still remains a challenging task. To address this problem, we introduce a novel framework based, in its core, on the idea of implicitly learning the joint distribution of RGB images and their corresponding camera poses using a discriminator network and adversarial learning. Our method allows not only to regress the camera pose from a single image, however, also offers a solely RGB-based solution for camera pose refinement using the discriminator network. Further, we show that our method can effectively be used to optimize the predicted camera poses and thus improve the localization accuracy. To this end, we validate our proposed method on the publicly available 7-Scenes dataset improving upon the results of direct camera pose regression methods.

Sergey Zakharov, Ivan Shugurov, Slobodan Ilic

In this paper we present a novel deep learning method for 3D object detection and 6D pose estimation from RGB images. Our method, named DPOD (Dense Pose Object Detector), estimates dense multi-class 2D-3D correspondence maps between an input image and available 3D models. Given the correspondences, a 6DoF pose is computed via PnP and RANSAC. An additional RGB pose refinement of the initial pose estimates is performed using a custom deep learning-based refinement scheme. Our results and comparison to a vast number of related works demonstrate that a large number of correspondences is beneficial for obtaining high-quality 6D poses both before and after refinement. Unlike other methods that mainly use real data for training and do not train on synthetic renderings, we perform evaluation on both synthetic and real training data demonstrating superior results before and after refinement when compared to all recent detectors. While being precise, the presented approach is still real-time capable.

Tolga Birdal, Benjamin Busam, Nassir Navab et al.

We present a novel and effective method for detecting 3D primitives in cluttered, unorganized point clouds, without axillary segmentation or type specification. We consider the quadric surfaces for encapsulating the basic building blocks of our environments - planes, spheres, ellipsoids, cones or cylinders, in a unified fashion. Moreover, quadrics allow us to model higher degree of freedom shapes, such as hyperboloids or paraboloids that could be used in non-rigid settings. We begin by contributing two novel quadric fits targeting 3D point sets that are endowed with tangent space information. Based upon the idea of aligning the quadric gradients with the surface normals, our first formulation is exact and requires as low as four oriented points. The second fit approximates the first, and reduces the computational effort. We theoretically analyze these fits with rigor, and give algebraic and geometric arguments. Next, by re-parameterizing the solution, we devise a new local Hough voting scheme on the null-space coefficients that is combined with RANSAC, reducing the complexity from $O(N^4)$ to $O(N^3)$ (three points). To the best of our knowledge, this is the first method capable of performing a generic cross-type multi-object primitive detection in difficult scenes without segmentation. Our extensive qualitative and quantitative results show that our method is efficient and flexible, as well as being accurate.

Benjamin Planche, Sergey Zakharov, Ziyan Wu et al.

While convolutional neural networks are dominating the field of computer vision, one usually does not have access to the large amount of domain-relevant data needed for their training. It thus became common to use available synthetic samples along domain adaptation schemes to prepare algorithms for the target domain. Tackling this problem from a different angle, we introduce a pipeline to map unseen target samples into the synthetic domain used to train task-specific methods. Denoising the data and retaining only the features these recognition algorithms are familiar with, our solution greatly improves their performance. As this mapping is easier to learn than the opposite one (ie to learn to generate realistic features to augment the source samples), we demonstrate how our whole solution can be trained purely on augmented synthetic data, and still perform better than methods trained with domain-relevant information (eg real images or realistic textures for the 3D models). Applying our approach to object recognition from texture-less CAD data, we present a custom generative network which fully utilizes the purely geometrical information to learn robust features and achieve a more refined mapping for unseen color images.

Haowen Deng, Tolga Birdal, Slobodan Ilic

We present PPF-FoldNet for unsupervised learning of 3D local descriptors on pure point cloud geometry. Based on the folding-based auto-encoding of well known point pair features, PPF-FoldNet offers many desirable properties: it necessitates neither supervision, nor a sensitive local reference frame, benefits from point-set sparsity, is end-to-end, fast, and can extract powerful rotation invariant descriptors. Thanks to a novel feature visualization, its evolution can be monitored to provide interpretable insights. Our extensive experiments demonstrate that despite having six degree-of-freedom invariance and lack of training labels, our network achieves state of the art results in standard benchmark datasets and outperforms its competitors when rotations and varying point densities are present. PPF-FoldNet achieves $9\%$ higher recall on standard benchmarks, $23\%$ higher recall when rotations are introduced into the same datasets and finally, a margin of $>35\%$ is attained when point density is significantly decreased.

Tolga Birdal, Umut Şimşekli, M. Onur Eken et al.

We introduce Tempered Geodesic Markov Chain Monte Carlo (TG-MCMC) algorithm for initializing pose graph optimization problems, arising in various scenarios such as SFM (structure from motion) or SLAM (simultaneous localization and mapping). TG-MCMC is first of its kind as it unites asymptotically global non-convex optimization on the spherical manifold of quaternions with posterior sampling, in order to provide both reliable initial poses and uncertainty estimates that are informative about the quality of individual solutions. We devise rigorous theoretical convergence guarantees for our method and extensively evaluate it on synthetic and real benchmark datasets. Besides its elegance in formulation and theory, we show that our method is robust to missing data, noise and the estimated uncertainties capture intuitive properties of the data.

Mai Bui, Shadi Albarqouni, Slobodan Ilic et al.

Scene coordinate regression has become an essential part of current camera re-localization methods. Different versions, such as regression forests and deep learning methods, have been successfully applied to estimate the corresponding camera pose given a single input image. In this work, we propose to regress the scene coordinates pixel-wise for a given RGB image by using deep learning. Compared to the recent methods, which usually employ RANSAC to obtain a robust pose estimate from the established point correspondences, we propose to regress confidences of these correspondences, which allows us to immediately discard erroneous predictions and improve the initial pose estimates. Finally, the resulting confidences can be used to score initial pose hypothesis and aid in pose refinement, offering a generalized solution to solve this task.

Mai Bui, Sergey Zakharov, Shadi Albarqouni et al.

In this work, we propose a method for object recognition and pose estimation from depth images using convolutional neural networks. Previous methods addressing this problem rely on manifold learning to learn low dimensional viewpoint descriptors and employ them in a nearest neighbor search on an estimated descriptor space. In comparison we create an efficient multi-task learning framework combining manifold descriptor learning and pose regression. By combining the strengths of manifold learning using triplet loss and pose regression, we could either estimate the pose directly reducing the complexity compared to NN search, or use learned descriptor for the NN descriptor matching. By in depth experimental evaluation of the novel loss function we observed that the view descriptors learned by the network are much more discriminative resulting in almost 30% increase regarding relative pose accuracy compared to related works. On the other hand, regarding directly regressed poses we obtained important improvement compared to simple pose regression. By leveraging the advantages of both manifold learning and regression tasks, we are able to improve the current state-of-the-art for object recognition and pose retrieval that we demonstrate through in depth experimental evaluation.

Sergey Zakharov, Benjamin Planche, Ziyan Wu et al.

With the increasing availability of large databases of 3D CAD models, depth-based recognition methods can be trained on an uncountable number of synthetically rendered images. However, discrepancies with the real data acquired from various depth sensors still noticeably impede progress. Previous works adopted unsupervised approaches to generate more realistic depth data, but they all require real scans for training, even if unlabeled. This still represents a strong requirement, especially when considering real-life/industrial settings where real training images are hard or impossible to acquire, but texture-less 3D models are available. We thus propose a novel approach leveraging only CAD models to bridge the realism gap. Purely trained on synthetic data, playing against an extensive augmentation pipeline in an unsupervised manner, our generative adversarial network learns to effectively segment depth images and recover the clean synthetic-looking depth information even from partial occlusions. As our solution is not only fully decoupled from the real domains but also from the task-specific analytics, the pre-processed scans can be handed to any kind and number of recognition methods also trained on synthetic data. Through various experiments, we demonstrate how this simplifies their training and consistently enhances their performance, with results on par with the same methods trained on real data, and better than usual approaches doing the reverse mapping.

Tolga Birdal, Benjamin Busam, Nassir Navab et al.

This paper proposes a segmentation-free, automatic and efficient procedure to detect general geometric quadric forms in point clouds, where clutter and occlusions are inevitable. Our everyday world is dominated by man-made objects which are designed using 3D primitives (such as planes, cones, spheres, cylinders, etc.). These objects are also omnipresent in industrial environments. This gives rise to the possibility of abstracting 3D scenes through primitives, thereby positions these geometric forms as an integral part of perception and high level 3D scene understanding. As opposed to state-of-the-art, where a tailored algorithm treats each primitive type separately, we propose to encapsulate all types in a single robust detection procedure. At the center of our approach lies a closed form 3D quadric fit, operating in both primal & dual spaces and requiring as low as 4 oriented-points. Around this fit, we design a novel, local null-space voting strategy to reduce the 4-point case to 3. Voting is coupled with the famous RANSAC and makes our algorithm orders of magnitude faster than its conventional counterparts. This is the first method capable of performing a generic cross-type multi-object primitive detection in difficult scenes. Results on synthetic and real datasets support the validity of our method.

Haowen Deng, Tolga Birdal, Slobodan Ilic

We present PPFNet - Point Pair Feature NETwork for deeply learning a globally informed 3D local feature descriptor to find correspondences in unorganized point clouds. PPFNet learns local descriptors on pure geometry and is highly aware of the global context, an important cue in deep learning. Our 3D representation is computed as a collection of point-pair-features combined with the points and normals within a local vicinity. Our permutation invariant network design is inspired by PointNet and sets PPFNet to be ordering-free. As opposed to voxelization, our method is able to consume raw point clouds to exploit the full sparsity. PPFNet uses a novel $\textit{N-tuple}$ loss and architecture injecting the global information naturally into the local descriptor. It shows that context awareness also boosts the local feature representation. Qualitative and quantitative evaluations of our network suggest increased recall, improved robustness and invariance as well as a vital step in the 3D descriptor extraction performance.

Wadim Kehl, Fabian Manhardt, Federico Tombari et al.

We present a novel method for detecting 3D model instances and estimating their 6D poses from RGB data in a single shot. To this end, we extend the popular SSD paradigm to cover the full 6D pose space and train on synthetic model data only. Our approach competes or surpasses current state-of-the-art methods that leverage RGB-D data on multiple challenging datasets. Furthermore, our method produces these results at around 10Hz, which is many times faster than the related methods. For the sake of reproducibility, we make our trained networks and detection code publicly available.

Tolga Birdal, Slobodan Ilic

We present an efficient and automatic approach for accurate reconstruction of instances of big 3D objects from multiple, unorganized and unstructured point clouds, in presence of dynamic clutter and occlusions. In contrast to conventional scanning, where the background is assumed to be rather static, we aim at handling dynamic clutter where background drastically changes during the object scanning. Currently, it is tedious to solve this with available methods unless the object of interest is first segmented out from the rest of the scene. We address the problem by assuming the availability of a prior CAD model, roughly resembling the object to be reconstructed. This assumption almost always holds in applications such as industrial inspection or reverse engineering. With aid of this prior acting as a proxy, we propose a fully enhanced pipeline, capable of automatically detecting and segmenting the object of interest from scenes and creating a pose graph, online, with linear complexity. This allows initial scan alignment to the CAD model space, which is then refined without the CAD constraint to fully recover a high fidelity 3D reconstruction, accurate up to the sensor noise level. We also contribute a novel object detection method, local implicit shape models (LISM) and give a fast verification scheme. We evaluate our method on multiple datasets, demonstrating the ability to accurately reconstruct objects from small sizes up to $125m^3$.

Wadim Kehl, Tobias Holl, Federico Tombari et al.

Volume-based reconstruction is usually expensive both in terms of memory consumption and runtime. Especially for sparse geometric structures, volumetric representations produce a huge computational overhead. We present an efficient way to fuse range data via a variational Octree-based minimization approach by taking the actual range data geometry into account. We transform the data into Octree-based truncated signed distance fields and show how the optimization can be conducted on the newly created structures. The main challenge is to uphold speed and a low memory footprint without sacrificing the solutions' accuracy during optimization. We explain how to dynamically adjust the optimizer's geometric structure via joining/splitting of Octree nodes and how to define the operators. We evaluate on various datasets and outline the suitability in terms of performance and geometric accuracy.

Wadim Kehl, Federico Tombari, Nassir Navab et al.

We present a scalable method for detecting objects and estimating their 3D poses in RGB-D data. To this end, we rely on an efficient representation of object views and employ hashing techniques to match these views against the input frame in a scalable way. While a similar approach already exists for 2D detection, we show how to extend it to estimate the 3D pose of the detected objects. In particular, we explore different hashing strategies and identify the one which is more suitable to our problem. We show empirically that the complexity of our method is sublinear with the number of objects and we enable detection and pose estimation of many 3D objects with high accuracy while outperforming the state-of-the-art in terms of runtime.

Wadim Kehl, Fausto Milletari, Federico Tombari et al.

We present a 3D object detection method that uses regressed descriptors of locally-sampled RGB-D patches for 6D vote casting. For regression, we employ a convolutional auto-encoder that has been trained on a large collection of random local patches. During testing, scene patch descriptors are matched against a database of synthetic model view patches and cast 6D object votes which are subsequently filtered to refined hypotheses. We evaluate on three datasets to show that our method generalizes well to previously unseen input data, delivers robust detection results that compete with and surpass the state-of-the-art while being scalable in the number of objects.

Hirokatsu Kataoka, Masaki Hayashi, Kenji Iwata et al.

In this paper, we propose a framework for recognizing human activities that uses only in-topic dominant codewords and a mixture of intertopic vectors. Latent Dirichlet allocation (LDA) is used to develop approximations of human motion primitives; these are mid-level representations, and they adaptively integrate dominant vectors when classifying human activities. In LDA topic modeling, action videos (documents) are represented by a bag-of-words (input from a dictionary), and these are based on improved dense trajectories. The output topics correspond to human motion primitives, such as finger moving or subtle leg motion. We eliminate the impurities, such as missed tracking or changing light conditions, in each motion primitive. The assembled vector of motion primitives is an improved representation of the action. We demonstrate our method on four different datasets.