Rares Ambrus

Prafull Sharma, Ayush Tewari, Yilun Du et al. · mit

We present a method to map 2D image observations of a scene to a persistent 3D scene representation, enabling novel view synthesis and disentangled representation of the movable and immovable components of the scene. Motivated by the bird's-eye-view (BEV) representation commonly used in vision and robotics, we propose conditional neural groundplans, ground-aligned 2D feature grids, as persistent and memory-efficient scene representations. Our method is trained self-supervised from unlabeled multi-view observations using differentiable rendering, and learns to complete geometry and appearance of occluded regions. In addition, we show that we can leverage multi-view videos at training time to learn to separately reconstruct static and movable components of the scene from a single image at test time. The ability to separately reconstruct movable objects enables a variety of downstream tasks using simple heuristics, such as extraction of object-centric 3D representations, novel view synthesis, instance-level segmentation, 3D bounding box prediction, and scene editing. This highlights the value of neural groundplans as a backbone for efficient 3D scene understanding models.

Muhammad Zubair Irshad, Sergey Zakharov, Rares Ambrus et al. · gatech

Our method studies the complex task of object-centric 3D understanding from a single RGB-D observation. As it is an ill-posed problem, existing methods suffer from low performance for both 3D shape and 6D pose and size estimation in complex multi-object scenarios with occlusions. We present ShAPO, a method for joint multi-object detection, 3D textured reconstruction, 6D object pose and size estimation. Key to ShAPO is a single-shot pipeline to regress shape, appearance and pose latent codes along with the masks of each object instance, which is then further refined in a sparse-to-dense fashion. A novel disentangled shape and appearance database of priors is first learned to embed objects in their respective shape and appearance space. We also propose a novel, octree-based differentiable optimization step, allowing us to further improve object shape, pose and appearance simultaneously under the learned latent space, in an analysis-by-synthesis fashion. Our novel joint implicit textured object representation allows us to accurately identify and reconstruct novel unseen objects without having access to their 3D meshes. Through extensive experiments, we show that our method, trained on simulated indoor scenes, accurately regresses the shape, appearance and pose of novel objects in the real-world with minimal fine-tuning. Our method significantly out-performs all baselines on the NOCS dataset with an 8% absolute improvement in mAP for 6D pose estimation. Project page: https://zubair-irshad.github.io/projects/ShAPO.html

Dian Chen, Jie Li, Vitor Guizilini et al.

3D object detection from visual sensors is a cornerstone capability of robotic systems. State-of-the-art methods focus on reasoning and decoding object bounding boxes from multi-view camera input. In this work we gain intuition from the integral role of multi-view consistency in 3D scene understanding and geometric learning. To this end, we introduce VEDet, a novel 3D object detection framework that exploits 3D multi-view geometry to improve localization through viewpoint awareness and equivariance. VEDet leverages a query-based transformer architecture and encodes the 3D scene by augmenting image features with positional encodings from their 3D perspective geometry. We design view-conditioned queries at the output level, which enables the generation of multiple virtual frames during training to learn viewpoint equivariance by enforcing multi-view consistency. The multi-view geometry injected at the input level as positional encodings and regularized at the loss level provides rich geometric cues for 3D object detection, leading to state-of-the-art performance on the nuScenes benchmark. The code and model are made available at https://github.com/TRI-ML/VEDet.

Muhammad Zubair Irshad, Sergey Zakharov, Katherine Liu et al. · gatech

Recent implicit neural representations have shown great results for novel view synthesis. However, existing methods require expensive per-scene optimization from many views hence limiting their application to real-world unbounded urban settings where the objects of interest or backgrounds are observed from very few views. To mitigate this challenge, we introduce a new approach called NeO 360, Neural fields for sparse view synthesis of outdoor scenes. NeO 360 is a generalizable method that reconstructs 360° scenes from a single or a few posed RGB images. The essence of our approach is in capturing the distribution of complex real-world outdoor 3D scenes and using a hybrid image-conditional triplanar representation that can be queried from any world point. Our representation combines the best of both voxel-based and bird's-eye-view (BEV) representations and is more effective and expressive than each. NeO 360's representation allows us to learn from a large collection of unbounded 3D scenes while offering generalizability to new views and novel scenes from as few as a single image during inference. We demonstrate our approach on the proposed challenging 360° unbounded dataset, called NeRDS 360, and show that NeO 360 outperforms state-of-the-art generalizable methods for novel view synthesis while also offering editing and composition capabilities. Project page: https://zubair-irshad.github.io/projects/neo360.html

Mayank Lunayach, Sergey Zakharov, Dian Chen et al. · gatech

In this work, we address the challenging task of 3D object recognition without the reliance on real-world 3D labeled data. Our goal is to predict the 3D shape, size, and 6D pose of objects within a single RGB-D image, operating at the category level and eliminating the need for CAD models during inference. While existing self-supervised methods have made strides in this field, they often suffer from inefficiencies arising from non-end-to-end processing, reliance on separate models for different object categories, and slow surface extraction during the training of implicit reconstruction models; thus hindering both the speed and real-world applicability of the 3D recognition process. Our proposed method leverages a multi-stage training pipeline, designed to efficiently transfer synthetic performance to the real-world domain. This approach is achieved through a combination of 2D and 3D supervised losses during the synthetic domain training, followed by the incorporation of 2D supervised and 3D self-supervised losses on real-world data in two additional learning stages. By adopting this comprehensive strategy, our method successfully overcomes the aforementioned limitations and outperforms existing self-supervised 6D pose and size estimation baselines on the NOCS test-set with a 16.4% absolute improvement in mAP for 6D pose estimation while running in near real-time at 5 Hz.

Adam W. Harley, Zhaoyuan Fang, Jie Li et al.

Building 3D perception systems for autonomous vehicles that do not rely on high-density LiDAR is a critical research problem because of the expense of LiDAR systems compared to cameras and other sensors. Recent research has developed a variety of camera-only methods, where features are differentiably "lifted" from the multi-camera images onto the 2D ground plane, yielding a "bird's eye view" (BEV) feature representation of the 3D space around the vehicle. This line of work has produced a variety of novel "lifting" methods, but we observe that other details in the training setups have shifted at the same time, making it unclear what really matters in top-performing methods. We also observe that using cameras alone is not a real-world constraint, considering that additional sensors like radar have been integrated into real vehicles for years already. In this paper, we first of all attempt to elucidate the high-impact factors in the design and training protocol of BEV perception models. We find that batch size and input resolution greatly affect performance, while lifting strategies have a more modest effect -- even a simple parameter-free lifter works well. Second, we demonstrate that radar data can provide a substantial boost to performance, helping to close the gap between camera-only and LiDAR-enabled systems. We analyze the radar usage details that lead to good performance, and invite the community to re-consider this commonly-neglected part of the sensor platform.

Tara Sadjadpour, Jie Li, Rares Ambrus et al.

Multi-object tracking is a cornerstone capability of any robotic system. The quality of tracking is largely dependent on the quality of the detector used. In many applications, such as autonomous vehicles, it is preferable to over-detect objects to avoid catastrophic outcomes due to missed detections. As a result, current state-of-the-art 3D detectors produce high rates of false-positives to ensure a low number of false-negatives. This can negatively affect tracking by making data association and track lifecycle management more challenging. Additionally, occasional false-negative detections due to difficult scenarios like occlusions can harm tracking performance. To address these issues in a unified framework, we propose to learn shape and spatio-temporal affinities between tracks and detections in consecutive frames. Our affinity provides a probabilistic matching that leads to robust data association, track lifecycle management, false-positive elimination, false-negative propagation, and sequential track confidence refinement. Though past 3D MOT approaches address a subset of components in this problem domain, we offer the first self-contained framework that addresses all these aspects of the 3D MOT problem. We quantitatively evaluate our method on the nuScenes tracking benchmark where we achieve 1st place amongst LiDAR-only trackers using CenterPoint detections. Our method estimates accurate and precise tracks, while decreasing the overall number of false-positive and false-negative tracks and increasing the number of true-positive tracks. We analyze our performance with 5 metrics, giving a comprehensive overview of our approach to indicate how our tracking framework may impact the ultimate goal of an autonomous mobile agent. We also present ablative experiments and qualitative results that demonstrate our framework's capabilities in complex scenarios.

Colton Stearns, Davis Rempe, Jie Li et al.

3D multi-object tracking aims to uniquely and consistently identify all mobile entities through time. Despite the rich spatiotemporal information available in this setting, current 3D tracking methods primarily rely on abstracted information and limited history, e.g. single-frame object bounding boxes. In this work, we develop a holistic representation of traffic scenes that leverages both spatial and temporal information of the actors in the scene. Specifically, we reformulate tracking as a spatiotemporal problem by representing tracked objects as sequences of time-stamped points and bounding boxes over a long temporal history. At each timestamp, we improve the location and motion estimates of our tracked objects through learned refinement over the full sequence of object history. By considering time and space jointly, our representation naturally encodes fundamental physical priors such as object permanence and consistency across time. Our spatiotemporal tracking framework achieves state-of-the-art performance on the Waymo and nuScenes benchmarks.

Vitor Guizilini, Igor Vasiljevic, Dian Chen et al.

Monocular depth estimation is scale-ambiguous, and thus requires scale supervision to produce metric predictions. Even so, the resulting models will be geometry-specific, with learned scales that cannot be directly transferred across domains. Because of that, recent works focus instead on relative depth, eschewing scale in favor of improved up-to-scale zero-shot transfer. In this work we introduce ZeroDepth, a novel monocular depth estimation framework capable of predicting metric scale for arbitrary test images from different domains and camera parameters. This is achieved by (i) the use of input-level geometric embeddings that enable the network to learn a scale prior over objects; and (ii) decoupling the encoder and decoder stages, via a variational latent representation that is conditioned on single frame information. We evaluated ZeroDepth targeting both outdoor (KITTI, DDAD, nuScenes) and indoor (NYUv2) benchmarks, and achieved a new state-of-the-art in both settings using the same pre-trained model, outperforming methods that train on in-domain data and require test-time scaling to produce metric estimates.

Vitor Guizilini, Rares Ambrus, Dian Chen et al.

Multi-frame depth estimation improves over single-frame approaches by also leveraging geometric relationships between images via feature matching, in addition to learning appearance-based features. In this paper we revisit feature matching for self-supervised monocular depth estimation, and propose a novel transformer architecture for cost volume generation. We use depth-discretized epipolar sampling to select matching candidates, and refine predictions through a series of self- and cross-attention layers. These layers sharpen the matching probability between pixel features, improving over standard similarity metrics prone to ambiguities and local minima. The refined cost volume is decoded into depth estimates, and the whole pipeline is trained end-to-end from videos using only a photometric objective. Experiments on the KITTI and DDAD datasets show that our DepthFormer architecture establishes a new state of the art in self-supervised monocular depth estimation, and is even competitive with highly specialized supervised single-frame architectures. We also show that our learned cross-attention network yields representations transferable across datasets, increasing the effectiveness of pre-training strategies. Project page: https://sites.google.com/tri.global/depthformer

Vitor Guizilini, Kuan-Hui Lee, Rares Ambrus et al.

Self-supervised monocular depth estimation enables robots to learn 3D perception from raw video streams. This scalable approach leverages projective geometry and ego-motion to learn via view synthesis, assuming the world is mostly static. Dynamic scenes, which are common in autonomous driving and human-robot interaction, violate this assumption. Therefore, they require modeling dynamic objects explicitly, for instance via estimating pixel-wise 3D motion, i.e. scene flow. However, the simultaneous self-supervised learning of depth and scene flow is ill-posed, as there are infinitely many combinations that result in the same 3D point. In this paper we propose DRAFT, a new method capable of jointly learning depth, optical flow, and scene flow by combining synthetic data with geometric self-supervision. Building upon the RAFT architecture, we learn optical flow as an intermediate task to bootstrap depth and scene flow learning via triangulation. Our algorithm also leverages temporal and geometric consistency losses across tasks to improve multi-task learning. Our DRAFT architecture simultaneously establishes a new state of the art in all three tasks in the self-supervised monocular setting on the standard KITTI benchmark. Project page: https://sites.google.com/tri.global/draft.

Vitor Guizilini, Igor Vasiljevic, Jiading Fang et al.

Modern 3D computer vision leverages learning to boost geometric reasoning, mapping image data to classical structures such as cost volumes or epipolar constraints to improve matching. These architectures are specialized according to the particular problem, and thus require significant task-specific tuning, often leading to poor domain generalization performance. Recently, generalist Transformer architectures have achieved impressive results in tasks such as optical flow and depth estimation by encoding geometric priors as inputs rather than as enforced constraints. In this paper, we extend this idea and propose to learn an implicit, multi-view consistent scene representation, introducing a series of 3D data augmentation techniques as a geometric inductive prior to increase view diversity. We also show that introducing view synthesis as an auxiliary task further improves depth estimation. Our Depth Field Networks (DeFiNe) achieve state-of-the-art results in stereo and video depth estimation without explicit geometric constraints, and improve on zero-shot domain generalization by a wide margin.

Sergey Zakharov, Rares Ambrus, Vitor Guizilini et al.

Synthetic data is a scalable alternative to manual supervision, but it requires overcoming the sim-to-real domain gap. This discrepancy between virtual and real worlds is addressed by two seemingly opposed approaches: improving the realism of simulation or foregoing realism entirely via domain randomization. In this paper, we show that the recent progress in neural rendering enables a new unified approach we call Photo-realistic Neural Domain Randomization (PNDR). We propose to learn a composition of neural networks that acts as a physics-based ray tracer generating high-quality renderings from scene geometry alone. Our approach is modular, composed of different neural networks for materials, lighting, and rendering, thus enabling randomization of different key image generation components in a differentiable pipeline. Once trained, our method can be combined with other methods and used to generate photo-realistic image augmentations online and significantly more efficiently than via traditional ray-tracing. We demonstrate the usefulness of PNDR through two downstream tasks: 6D object detection and monocular depth estimation. Our experiments show that training with PNDR enables generalization to novel scenes and significantly outperforms the state of the art in terms of real-world transfer.

Tara Sadjadpour, Rares Ambrus, Jeannette Bohg

3D multi-object tracking (MOT) is essential for an autonomous mobile agent to safely navigate a scene. In order to maximize the perception capabilities of the autonomous agent, we aim to develop a 3D MOT framework that fuses camera and LiDAR sensor information. Building on our prior LiDAR-only work, ShaSTA, which models shape and spatio-temporal affinities for 3D MOT, we propose a novel camera-LiDAR fusion approach for learning affinities. At its core, this work proposes a fusion technique that generates a rich sensory signal incorporating information about depth and distant objects to enhance affinity estimation for improved data association, track lifecycle management, false-positive elimination, false-negative propagation, and track confidence score refinement. Our main contributions include a novel fusion approach for combining camera and LiDAR sensory signals to learn affinities, and a first-of-its-kind multimodal sequential track confidence refinement technique that fuses 2D and 3D detections. Additionally, we perform an ablative analysis on each fusion step to demonstrate the added benefits of incorporating the camera sensor, particular for small, distant objects that tend to suffer from the depth-sensing limits and sparsity of LiDAR sensors. In sum, our technique achieves state-of-the-art performance on the nuScenes benchmark amongst multimodal 3D MOT algorithms using CenterPoint detections.

Arkadeep Narayan Chaudhury, Igor Vasiljevic, Sergey Zakharov et al.

Synthesizing accurate geometry and photo-realistic appearance of small scenes is an active area of research with compelling use cases in gaming, virtual reality, robotic-manipulation, autonomous driving, convenient product capture, and consumer-level photography. When applying scene geometry and appearance estimation techniques to robotics, we found that the narrow cone of possible viewpoints due to the limited range of robot motion and scene clutter caused current estimation techniques to produce poor quality estimates or even fail. On the other hand, in robotic applications, dense metric depth can often be measured directly using stereo and illumination can be controlled. Depth can provide a good initial estimate of the object geometry to improve reconstruction, while multi-illumination images can facilitate relighting. In this work we demonstrate a method to incorporate dense metric depth into the training of neural 3D representations and address an artifact observed while jointly refining geometry and appearance by disambiguating between texture and geometry edges. We also discuss a multi-flash stereo camera system developed to capture the necessary data for our pipeline and show results on relighting and view synthesis with a few training views.

Vitor Guizilini, Pavel Tokmakov, Achal Dave et al.

3D reconstruction from a single image is a long-standing problem in computer vision. Learning-based methods address its inherent scale ambiguity by leveraging increasingly large labeled and unlabeled datasets, to produce geometric priors capable of generating accurate predictions across domains. As a result, state of the art approaches show impressive performance in zero-shot relative and metric depth estimation. Recently, diffusion models have exhibited remarkable scalability and generalizable properties in their learned representations. However, because these models repurpose tools originally designed for image generation, they can only operate on dense ground-truth, which is not available for most depth labels, especially in real-world settings. In this paper we present GRIN, an efficient diffusion model designed to ingest sparse unstructured training data. We use image features with 3D geometric positional encodings to condition the diffusion process both globally and locally, generating depth predictions at a pixel-level. With comprehensive experiments across eight indoor and outdoor datasets, we show that GRIN establishes a new state of the art in zero-shot metric monocular depth estimation even when trained from scratch.

Takayuki Kanai, Igor Vasiljevic, Vitor Guizilini et al.

Autonomous vehicles and robots need to operate over a wide variety of scenarios in order to complete tasks efficiently and safely. Multi-camera self-supervised monocular depth estimation from videos is a promising way to reason about the environment, as it generates metrically scaled geometric predictions from visual data without requiring additional sensors. However, most works assume well-calibrated extrinsics to fully leverage this multi-camera setup, even though accurate and efficient calibration is still a challenging problem. In this work, we introduce a novel method for extrinsic calibration that builds upon the principles of self-supervised monocular depth and ego-motion learning. Our proposed curriculum learning strategy uses monocular depth and pose estimators with velocity supervision to estimate extrinsics, and then jointly learns extrinsic calibration along with depth and pose for a set of overlapping cameras rigidly attached to a moving vehicle. Experiments on a benchmark multi-camera dataset (DDAD) demonstrate that our method enables self-calibration in various scenes robustly and efficiently compared to a traditional vision-based pose estimation pipeline. Furthermore, we demonstrate the benefits of extrinsics self-calibration as a way to improve depth prediction via joint optimization.

Vitor Guizilini, Igor Vasiljevic, Jiading Fang et al.

Differentiable volumetric rendering is a powerful paradigm for 3D reconstruction and novel view synthesis. However, standard volume rendering approaches struggle with degenerate geometries in the case of limited viewpoint diversity, a common scenario in robotics applications. In this work, we propose to use the multi-view photometric objective from the self-supervised depth estimation literature as a geometric regularizer for volumetric rendering, significantly improving novel view synthesis without requiring additional information. Building upon this insight, we explore the explicit modeling of scene geometry using a generalist Transformer, jointly learning a radiance field as well as depth and light fields with a set of shared latent codes. We demonstrate that sharing geometric information across tasks is mutually beneficial, leading to improvements over single-task learning without an increase in network complexity. Our DeLiRa architecture achieves state-of-the-art results on the ScanNet benchmark, enabling high quality volumetric rendering as well as real-time novel view and depth synthesis in the limited viewpoint diversity setting.

Sergey Zakharov, Rares Ambrus, Katherine Liu et al.

Compact and accurate representations of 3D shapes are central to many perception and robotics tasks. State-of-the-art learning-based methods can reconstruct single objects but scale poorly to large datasets. We present a novel recursive implicit representation to efficiently and accurately encode large datasets of complex 3D shapes by recursively traversing an implicit octree in latent space. Our implicit Recursive Octree Auto-Decoder (ROAD) learns a hierarchically structured latent space enabling state-of-the-art reconstruction results at a compression ratio above 99%. We also propose an efficient curriculum learning scheme that naturally exploits the coarse-to-fine properties of the underlying octree spatial representation. We explore the scaling law relating latent space dimension, dataset size, and reconstruction accuracy, showing that increasing the latent space dimension is enough to scale to large shape datasets. Finally, we show that our learned latent space encodes a coarse-to-fine hierarchical structure yielding reusable latents across different levels of details, and we provide qualitative evidence of generalization to novel shapes outside the training set.

Adam W. Harley, Yang You, Xinglong Sun et al.

We introduce AllTracker: a model that estimates long-range point tracks by way of estimating the flow field between a query frame and every other frame of a video. Unlike existing point tracking methods, our approach delivers high-resolution and dense (all-pixel) correspondence fields, which can be visualized as flow maps. Unlike existing optical flow methods, our approach corresponds one frame to hundreds of subsequent frames, rather than just the next frame. We develop a new architecture for this task, blending techniques from existing work in optical flow and point tracking: the model performs iterative inference on low-resolution grids of correspondence estimates, propagating information spatially via 2D convolution layers, and propagating information temporally via pixel-aligned attention layers. The model is fast and parameter-efficient (16 million parameters), and delivers state-of-the-art point tracking accuracy at high resolution (i.e., tracking 768x1024 pixels, on a 40G GPU). A benefit of our design is that we can train jointly on optical flow datasets and point tracking datasets, and we find that doing so is crucial for top performance. We provide an extensive ablation study on our architecture details and training recipe, making it clear which details matter most. Our code and model weights are available at https://alltracker.github.io

Takuya Ikeda, Sergey Zakharov, Tianyi Ko et al. · gatech

This paper addresses the challenging problem of category-level pose estimation. Current state-of-the-art methods for this task face challenges when dealing with symmetric objects and when attempting to generalize to new environments solely through synthetic data training. In this work, we address these challenges by proposing a probabilistic model that relies on diffusion to estimate dense canonical maps crucial for recovering partial object shapes as well as establishing correspondences essential for pose estimation. Furthermore, we introduce critical components to enhance performance by leveraging the strength of the diffusion models with multi-modal input representations. We demonstrate the effectiveness of our method by testing it on a range of real datasets. Despite being trained solely on our generated synthetic data, our approach achieves state-of-the-art performance and unprecedented generalization qualities, outperforming baselines, even those specifically trained on the target domain.

Muhammad Zubair Irshad, Sergey Zakharov, Vitor Guizilini et al. · gatech

Neural fields excel in computer vision and robotics due to their ability to understand the 3D visual world such as inferring semantics, geometry, and dynamics. Given the capabilities of neural fields in densely representing a 3D scene from 2D images, we ask the question: Can we scale their self-supervised pretraining, specifically using masked autoencoders, to generate effective 3D representations from posed RGB images. Owing to the astounding success of extending transformers to novel data modalities, we employ standard 3D Vision Transformers to suit the unique formulation of NeRFs. We leverage NeRF's volumetric grid as a dense input to the transformer, contrasting it with other 3D representations such as pointclouds where the information density can be uneven, and the representation is irregular. Due to the difficulty of applying masked autoencoders to an implicit representation, such as NeRF, we opt for extracting an explicit representation that canonicalizes scenes across domains by employing the camera trajectory for sampling. Our goal is made possible by masking random patches from NeRF's radiance and density grid and employing a standard 3D Swin Transformer to reconstruct the masked patches. In doing so, the model can learn the semantic and spatial structure of complete scenes. We pretrain this representation at scale on our proposed curated posed-RGB data, totaling over 1.8 million images. Once pretrained, the encoder is used for effective 3D transfer learning. Our novel self-supervised pretraining for NeRFs, NeRF-MAE, scales remarkably well and improves performance on various challenging 3D tasks. Utilizing unlabeled posed 2D data for pretraining, NeRF-MAE significantly outperforms self-supervised 3D pretraining and NeRF scene understanding baselines on Front3D and ScanNet datasets with an absolute performance improvement of over 20% AP50 and 8% AP25 for 3D object detection.

Muhammad Zubair Irshad, Mauro Comi, Yen-Chen Lin et al. · gatech

Neural Fields have emerged as a transformative approach for 3D scene representation in computer vision and robotics, enabling accurate inference of geometry, 3D semantics, and dynamics from posed 2D data. Leveraging differentiable rendering, Neural Fields encompass both continuous implicit and explicit neural representations enabling high-fidelity 3D reconstruction, integration of multi-modal sensor data, and generation of novel viewpoints. This survey explores their applications in robotics, emphasizing their potential to enhance perception, planning, and control. Their compactness, memory efficiency, and differentiability, along with seamless integration with foundation and generative models, make them ideal for real-time applications, improving robot adaptability and decision-making. This paper provides a thorough review of Neural Fields in robotics, categorizing applications across various domains and evaluating their strengths and limitations, based on over 200 papers. First, we present four key Neural Fields frameworks: Occupancy Networks, Signed Distance Fields, Neural Radiance Fields, and Gaussian Splatting. Second, we detail Neural Fields' applications in five major robotics domains: pose estimation, manipulation, navigation, physics, and autonomous driving, highlighting key works and discussing takeaways and open challenges. Finally, we outline the current limitations of Neural Fields in robotics and propose promising directions for future research. Project page: https://robonerf.github.io

Chen Xu, Tony Khuong Nguyen, Emma Dixon et al.

Recent years have witnessed impressive robotic manipulation systems driven by advances in imitation learning and generative modeling, such as diffusion- and flow-based approaches. As robot policy performance increases, so does the complexity and time horizon of achievable tasks, inducing unexpected and diverse failure modes that are difficult to predict a priori. To enable trustworthy policy deployment in safety-critical human environments, reliable runtime failure detection becomes important during policy inference. However, most existing failure detection approaches rely on prior knowledge of failure modes and require failure data during training, which imposes a significant challenge in practicality and scalability. In response to these limitations, we present FAIL-Detect, a modular two-stage approach for failure detection in imitation learning-based robotic manipulation. To accurately identify failures from successful training data alone, we frame the problem as sequential out-of-distribution (OOD) detection. We first distill policy inputs and outputs into scalar signals that correlate with policy failures and capture epistemic uncertainty. FAIL-Detect then employs conformal prediction (CP) as a versatile framework for uncertainty quantification with statistical guarantees. Empirically, we thoroughly investigate both learned and post-hoc scalar signal candidates on diverse robotic manipulation tasks. Our experiments show learned signals to be mostly consistently effective, particularly when using our novel flow-based density estimator. Furthermore, our method detects failures more accurately and faster than state-of-the-art (SOTA) failure detection baselines. These results highlight the potential of FAIL-Detect to enhance the safety and reliability of imitation learning-based robotic systems as they progress toward real-world deployment.

Jiading Fang, Xiangshan Tan, Shengjie Lin et al.

If robots are to work effectively alongside people, they must be able to interpret natural language references to objects in their 3D environment. Understanding 3D referring expressions is challenging -- it requires the ability to both parse the 3D structure of the scene and correctly ground free-form language in the presence of distraction and clutter. We introduce Transcrib3D, an approach that brings together 3D detection methods and the emergent reasoning capabilities of large language models (LLMs). Transcrib3D uses text as the unifying medium, which allows us to sidestep the need to learn shared representations connecting multi-modal inputs, which would require massive amounts of annotated 3D data. As a demonstration of its effectiveness, Transcrib3D achieves state-of-the-art results on 3D reference resolution benchmarks, with a great leap in performance from previous multi-modality baselines. To improve upon zero-shot performance and facilitate local deployment on edge computers and robots, we propose self-correction for fine-tuning that trains smaller models, resulting in performance close to that of large models. We show that our method enables a real robot to perform pick-and-place tasks given queries that contain challenging referring expressions. Project site is at https://ripl.github.io/Transcrib3D.

Vitor Guizilini, Muhammad Zubair Irshad, Dian Chen et al. · gatech

Current methods for 3D scene reconstruction from sparse posed images employ intermediate 3D representations such as neural fields, voxel grids, or 3D Gaussians, to achieve multi-view consistent scene appearance and geometry. In this paper we introduce MVGD, a diffusion-based architecture capable of direct pixel-level generation of images and depth maps from novel viewpoints, given an arbitrary number of input views. Our method uses raymap conditioning to both augment visual features with spatial information from different viewpoints, as well as to guide the generation of images and depth maps from novel views. A key aspect of our approach is the multi-task generation of images and depth maps, using learnable task embeddings to guide the diffusion process towards specific modalities. We train this model on a collection of more than 60 million multi-view samples from publicly available datasets, and propose techniques to enable efficient and consistent learning in such diverse conditions. We also propose a novel strategy that enables the efficient training of larger models by incrementally fine-tuning smaller ones, with promising scaling behavior. Through extensive experiments, we report state-of-the-art results in multiple novel view synthesis benchmarks, as well as multi-view stereo and video depth estimation.

Shun Iwase, Katherine Liu, Vitor Guizilini et al.

We present a 3D scene completion method that recovers the complete geometry of multiple unseen objects in complex scenes from a single RGB-D image. Despite notable advancements in single-object 3D shape completion, high-quality reconstructions in highly cluttered real-world multi-object scenes remains a challenge. To address this issue, we propose OctMAE, an architecture that leverages an Octree U-Net and a latent 3D MAE to achieve high-quality and near real-time multi-object scene completion through both local and global geometric reasoning. Because a naive 3D MAE can be computationally intractable and memory intensive even in the latent space, we introduce a novel occlusion masking strategy and adopt 3D rotary embeddings, which significantly improves the runtime and scene completion quality. To generalize to a wide range of objects in diverse scenes, we create a large-scale photorealistic dataset, featuring a diverse set of 12K 3D object models from the Objaverse dataset which are rendered in multi-object scenes with physics-based positioning. Our method outperforms the current state-of-the-art on both synthetic and real-world datasets and demonstrates a strong zero-shot capability.

Yinshuang Xu, Dian Chen, Katherine Liu et al.

Incorporating inductive bias by embedding geometric entities (such as rays) as input has proven successful in multi-view learning. However, the methods adopting this technique typically lack equivariance, which is crucial for effective 3D learning. Equivariance serves as a valuable inductive prior, aiding in the generation of robust multi-view features for 3D scene understanding. In this paper, we explore the application of equivariant multi-view learning to depth estimation, not only recognizing its significance for computer vision and robotics but also addressing the limitations of previous research. Most prior studies have either overlooked equivariance in this setting or achieved only approximate equivariance through data augmentation, which often leads to inconsistencies across different reference frames. To address this issue, we propose to embed $SE(3)$ equivariance into the Perceiver IO architecture. We employ Spherical Harmonics for positional encoding to ensure 3D rotation equivariance, and develop a specialized equivariant encoder and decoder within the Perceiver IO architecture. To validate our model, we applied it to the task of stereo depth estimation, achieving state of the art results on real-world datasets without explicit geometric constraints or extensive data augmentation.

Shun Iwase, Zubair Irshad, Katherine Liu et al. · gatech

Robotic grasping is a cornerstone capability of embodied systems. Many methods directly output grasps from partial information without modeling the geometry of the scene, leading to suboptimal motion and even collisions. To address these issues, we introduce ZeroGrasp, a novel framework that simultaneously performs 3D reconstruction and grasp pose prediction in near real-time. A key insight of our method is that occlusion reasoning and modeling the spatial relationships between objects is beneficial for both accurate reconstruction and grasping. We couple our method with a novel large-scale synthetic dataset, which comprises 1M photo-realistic images, high-resolution 3D reconstructions and 11.3B physically-valid grasp pose annotations for 12K objects from the Objaverse-LVIS dataset. We evaluate ZeroGrasp on the GraspNet-1B benchmark as well as through real-world robot experiments. ZeroGrasp achieves state-of-the-art performance and generalizes to novel real-world objects by leveraging synthetic data.

Keunwoo Peter Yu, Achal Dave, Rares Ambrus et al.

Recent advances in vision-language models (VLMs) have shown great promise in connecting images and text, but extending these models to long videos remains challenging due to the rapid growth in token counts. Models that compress videos by local aggregation in time or space have become popular for handling long-form inputs; however, these pooling-based projectors sacrifice the benefits of fixed-length representations that are crucial for streaming and efficient video understanding. We introduce $\texttt{Espresso}$, a new architecture that separately compresses spatial and temporal features into fixed-length sequences. $\texttt{Espresso}$ enables efficient video encoding while maintaining strong long-form reasoning capabilities. Experiments show that fixed-length compression combined with segment-wise processing offers a scalable and competitive alternative to pooling-based approaches. Our results demonstrate that fixed-length projectors, when properly designed and trained, remain a viable foundation for video-language modeling.

Katherine Liu, Sergey Zakharov, Dian Chen et al.

We would like to estimate the pose and full shape of an object from a single observation, without assuming known 3D model or category. In this work, we propose OmniShape, the first method of its kind to enable probabilistic pose and shape estimation. OmniShape is based on the key insight that shape completion can be decoupled into two multi-modal distributions: one capturing how measurements project into a normalized object reference frame defined by the dataset and the other modelling a prior over object geometries represented as triplanar neural fields. By training separate conditional diffusion models for these two distributions, we enable sampling multiple hypotheses from the joint pose and shape distribution. OmniShape demonstrates compelling performance on challenging real world datasets. Project website: https://tri-ml.github.io/omnishape

Takuya Ikeda, Sergey Zakharov, Muhammad Zubair Irshad et al. · gatech

We present a novel method for 6-DoF object tracking and high-quality 3D reconstruction from monocular RGBD video. Existing methods, while achieving impressive results, often struggle with complex objects, particularly those exhibiting symmetry, intricate geometry or complex appearance. To bridge these gaps, we introduce an adaptive method that combines 3D Gaussian Splatting, hybrid geometry/appearance tracking, and key frame selection to achieve robust tracking and accurate reconstructions across a diverse range of objects. Additionally, we present a benchmark covering these challenging object classes, providing high-quality annotations for evaluating both tracking and reconstruction performance. Our approach demonstrates strong capabilities in recovering high-fidelity object meshes, setting a new standard for single-sensor 3D reconstruction in open-world environments.

Sergey Zakharov, Katherine Liu, Adrien Gaidon et al.

The common trade-offs of state-of-the-art methods for multi-shape representation (a single model "packing" multiple objects) involve trading modeling accuracy against memory and storage. We show how to encode multiple shapes represented as continuous neural fields with a higher degree of precision than previously possible and with low memory usage. Key to our approach is a recursive hierarchical formulation that exploits object self-similarity, leading to a highly compressed and efficient shape latent space. Thanks to the recursive formulation, our method supports spatial and global-to-local latent feature fusion without needing to initialize and maintain auxiliary data structures, while still allowing for continuous field queries to enable applications such as raytracing. In experiments on a set of diverse datasets, we provide compelling qualitative results and demonstrate state-of-the-art multi-scene reconstruction and compression results with a single network per dataset.

Matthew Kowal, Achal Dave, Rares Ambrus et al.

This paper studies the problem of concept-based interpretability of transformer representations for videos. Concretely, we seek to explain the decision-making process of video transformers based on high-level, spatiotemporal concepts that are automatically discovered. Prior research on concept-based interpretability has concentrated solely on image-level tasks. Comparatively, video models deal with the added temporal dimension, increasing complexity and posing challenges in identifying dynamic concepts over time. In this work, we systematically address these challenges by introducing the first Video Transformer Concept Discovery (VTCD) algorithm. To this end, we propose an efficient approach for unsupervised identification of units of video transformer representations - concepts, and ranking their importance to the output of a model. The resulting concepts are highly interpretable, revealing spatio-temporal reasoning mechanisms and object-centric representations in unstructured video models. Performing this analysis jointly over a diverse set of supervised and self-supervised representations, we discover that some of these mechanism are universal in video transformers. Finally, we show that VTCD can be used for fine-grained action recognition and video object segmentation.

Jiading Fang, Shengjie Lin, Igor Vasiljevic et al.

A practical benefit of implicit visual representations like Neural Radiance Fields (NeRFs) is their memory efficiency: large scenes can be efficiently stored and shared as small neural nets instead of collections of images. However, operating on these implicit visual data structures requires extending classical image-based vision techniques (e.g., registration, blending) from image sets to neural fields. Towards this goal, we propose NeRFuser, a novel architecture for NeRF registration and blending that assumes only access to pre-generated NeRFs, and not the potentially large sets of images used to generate them. We propose registration from re-rendering, a technique to infer the transformation between NeRFs based on images synthesized from individual NeRFs. For blending, we propose sample-based inverse distance weighting to blend visual information at the ray-sample level. We evaluate NeRFuser on public benchmarks and a self-collected object-centric indoor dataset, showing the robustness of our method, including to views that are challenging to render from the individual source NeRFs.

Jiading Fang, Igor Vasiljevic, Vitor Guizilini et al.

Camera calibration is integral to robotics and computer vision algorithms that seek to infer geometric properties of the scene from visual input streams. In practice, calibration is a laborious procedure requiring specialized data collection and careful tuning. This process must be repeated whenever the parameters of the camera change, which can be a frequent occurrence for mobile robots and autonomous vehicles. In contrast, self-supervised depth and ego-motion estimation approaches can bypass explicit calibration by inferring per-frame projection models that optimize a view synthesis objective. In this paper, we extend this approach to explicitly calibrate a wide range of cameras from raw videos in the wild. We propose a learning algorithm to regress per-sequence calibration parameters using an efficient family of general camera models. Our procedure achieves self-calibration results with sub-pixel reprojection error, outperforming other learning-based methods. We validate our approach on a wide variety of camera geometries, including perspective, fisheye, and catadioptric. Finally, we show that our approach leads to improvements in the downstream task of depth estimation, achieving state-of-the-art results on the EuRoC dataset with greater computational efficiency than contemporary methods.

Aditya Ganeshan, Alexis Vallet, Yasunori Kudo et al.

Deep learning models for semantic segmentation rely on expensive, large-scale, manually annotated datasets. Labelling is a tedious process that can take hours per image. Automatically annotating video sequences by propagating sparsely labeled frames through time is a more scalable alternative. In this work, we propose a novel label propagation method, termed Warp-Refine Propagation, that combines semantic cues with geometric cues to efficiently auto-label videos. Our method learns to refine geometrically-warped labels and infuse them with learned semantic priors in a semi-supervised setting by leveraging cycle consistency across time. We quantitatively show that our method improves label-propagation by a noteworthy margin of 13.1 mIoU on the ApolloScape dataset. Furthermore, by training with the auto-labelled frames, we achieve competitive results on three semantic-segmentation benchmarks, improving the state-of-the-art by a large margin of 1.8 and 3.61 mIoU on NYU-V2 and KITTI, while matching the current best results on Cityscapes.

Dennis Park, Rares Ambrus, Vitor Guizilini et al.

Recent progress in 3D object detection from single images leverages monocular depth estimation as a way to produce 3D pointclouds, turning cameras into pseudo-lidar sensors. These two-stage detectors improve with the accuracy of the intermediate depth estimation network, which can itself be improved without manual labels via large-scale self-supervised learning. However, they tend to suffer from overfitting more than end-to-end methods, are more complex, and the gap with similar lidar-based detectors remains significant. In this work, we propose an end-to-end, single stage, monocular 3D object detector, DD3D, that can benefit from depth pre-training like pseudo-lidar methods, but without their limitations. Our architecture is designed for effective information transfer between depth estimation and 3D detection, allowing us to scale with the amount of unlabeled pre-training data. Our method achieves state-of-the-art results on two challenging benchmarks, with 16.34% and 9.28% AP for Cars and Pedestrians (respectively) on the KITTI-3D benchmark, and 41.5% mAP on NuScenes.

Vitor Guizilini, Igor Vasiljevic, Rares Ambrus et al.

Self-supervised monocular depth and ego-motion estimation is a promising approach to replace or supplement expensive depth sensors such as LiDAR for robotics applications like autonomous driving. However, most research in this area focuses on a single monocular camera or stereo pairs that cover only a fraction of the scene around the vehicle. In this work, we extend monocular self-supervised depth and ego-motion estimation to large-baseline multi-camera rigs. Using generalized spatio-temporal contexts, pose consistency constraints, and carefully designed photometric loss masking, we learn a single network generating dense, consistent, and scale-aware point clouds that cover the same full surround 360 degree field of view as a typical LiDAR scanner. We also propose a new scale-consistent evaluation metric more suitable to multi-camera settings. Experiments on two challenging benchmarks illustrate the benefits of our approach over strong baselines.

Vitor Guizilini, Jie Li, Rares Ambrus et al.

Simulators can efficiently generate large amounts of labeled synthetic data with perfect supervision for hard-to-label tasks like semantic segmentation. However, they introduce a domain gap that severely hurts real-world performance. We propose to use self-supervised monocular depth estimation as a proxy task to bridge this gap and improve sim-to-real unsupervised domain adaptation (UDA). Our Geometric Unsupervised Domain Adaptation method (GUDA) learns a domain-invariant representation via a multi-task objective combining synthetic semantic supervision with real-world geometric constraints on videos. GUDA establishes a new state of the art in UDA for semantic segmentation on three benchmarks, outperforming methods that use domain adversarial learning, self-training, or other self-supervised proxy tasks. Furthermore, we show that our method scales well with the quality and quantity of synthetic data while also improving depth prediction.

Vitor Guizilini, Rares Ambrus, Wolfram Burgard et al.

Estimating scene geometry from data obtained with cost-effective sensors is key for robots and self-driving cars. In this paper, we study the problem of predicting dense depth from a single RGB image (monodepth) with optional sparse measurements from low-cost active depth sensors. We introduce Sparse Auxiliary Networks (SANs), a new module enabling monodepth networks to perform both the tasks of depth prediction and completion, depending on whether only RGB images or also sparse point clouds are available at inference time. First, we decouple the image and depth map encoding stages using sparse convolutions to process only the valid depth map pixels. Second, we inject this information, when available, into the skip connections of the depth prediction network, augmenting its features. Through extensive experimental analysis on one indoor (NYUv2) and two outdoor (KITTI and DDAD) benchmarks, we demonstrate that our proposed SAN architecture is able to simultaneously learn both tasks, while achieving a new state of the art in depth prediction by a significant margin.

Rares Ambrus, Vitor Guizilini, Naveen Kuppuswamy et al.

Fluid-filled soft visuotactile sensors such as the Soft-bubbles alleviate key challenges for robust manipulation, as they enable reliable grasps along with the ability to obtain high-resolution sensory feedback on contact geometry and forces. Although they are simple in construction, their utility has been limited due to size constraints introduced by enclosed custom IR/depth imaging sensors to directly measure surface deformations. Towards mitigating this limitation, we investigate the application of state-of-the-art monocular depth estimation to infer dense internal (tactile) depth maps directly from the internal single small IR imaging sensor. Through real-world experiments, we show that deep networks typically used for long-range depth estimation (1-100m) can be effectively trained for precise predictions at a much shorter range (1-100mm) inside a mostly textureless deformable fluid-filled sensor. We propose a simple supervised learning process to train an object-agnostic network requiring less than 10 random poses in contact for less than 10 seconds for a small set of diverse objects (mug, wine glass, box, and fingers in our experiments). We show that our approach is sample-efficient, accurate, and generalizes across different objects and sensor configurations unseen at training time. Finally, we discuss the implications of our approach for the design of soft visuotactile sensors and grippers.

Hsu-kuang Chiu, Jie Li, Rares Ambrus et al.

Multi-object tracking is an important ability for an autonomous vehicle to safely navigate a traffic scene. Current state-of-the-art follows the tracking-by-detection paradigm where existing tracks are associated with detected objects through some distance metric. The key challenges to increase tracking accuracy lie in data association and track life cycle management. We propose a probabilistic, multi-modal, multi-object tracking system consisting of different trainable modules to provide robust and data-driven tracking results. First, we learn how to fuse features from 2D images and 3D LiDAR point clouds to capture the appearance and geometric information of an object. Second, we propose to learn a metric that combines the Mahalanobis and feature distances when comparing a track and a new detection in data association. And third, we propose to learn when to initialize a track from an unmatched object detection. Through extensive quantitative and qualitative results, we show that when using the same object detectors our method outperforms state-of-the-art approaches on the NuScenes and KITTI datasets.

Andreas Bühler, Adrien Gaidon, Andrei Cramariuc et al.

Safe autonomous driving requires robust detection of other traffic participants. However, robust does not mean perfect, and safe systems typically minimize missed detections at the expense of a higher false positive rate. This results in conservative and yet potentially dangerous behavior such as avoiding imaginary obstacles. In the context of behavioral cloning, perceptual errors at training time can lead to learning difficulties or wrong policies, as expert demonstrations might be inconsistent with the perceived world state. In this work, we propose a behavioral cloning approach that can safely leverage imperfect perception without being conservative. Our core contribution is a novel representation of perceptual uncertainty for learning to plan. We propose a new probabilistic birds-eye-view semantic grid to encode the noisy output of object perception systems. We then leverage expert demonstrations to learn an imitative driving policy using this probabilistic representation. Using the CARLA simulator, we show that our approach can safely overcome critical false positives that would otherwise lead to catastrophic failures or conservative behavior.

Igor Vasiljevic, Vitor Guizilini, Rares Ambrus et al.

Self-supervised learning has emerged as a powerful tool for depth and ego-motion estimation, leading to state-of-the-art results on benchmark datasets. However, one significant limitation shared by current methods is the assumption of a known parametric camera model -- usually the standard pinhole geometry -- leading to failure when applied to imaging systems that deviate significantly from this assumption (e.g., catadioptric cameras or underwater imaging). In this work, we show that self-supervision can be used to learn accurate depth and ego-motion estimation without prior knowledge of the camera model. Inspired by the geometric model of Grossberg and Nayar, we introduce Neural Ray Surfaces (NRS), convolutional networks that represent pixel-wise projection rays, approximating a wide range of cameras. NRS are fully differentiable and can be learned end-to-end from unlabeled raw videos. We demonstrate the use of NRS for self-supervised learning of visual odometry and depth estimation from raw videos obtained using a wide variety of camera systems, including pinhole, fisheye, and catadioptric.

Vitor Guizilini, Rui Hou, Jie Li et al.

Self-supervised learning is showing great promise for monocular depth estimation, using geometry as the only source of supervision. Depth networks are indeed capable of learning representations that relate visual appearance to 3D properties by implicitly leveraging category-level patterns. In this work we investigate how to leverage more directly this semantic structure to guide geometric representation learning, while remaining in the self-supervised regime. Instead of using semantic labels and proxy losses in a multi-task approach, we propose a new architecture leveraging fixed pretrained semantic segmentation networks to guide self-supervised representation learning via pixel-adaptive convolutions. Furthermore, we propose a two-stage training process to overcome a common semantic bias on dynamic objects via resampling. Our method improves upon the state of the art for self-supervised monocular depth prediction over all pixels, fine-grained details, and per semantic categories.

Jiexiong Tang, Hanme Kim, Vitor Guizilini et al.

Identifying salient points in images is a crucial component for visual odometry, Structure-from-Motion or SLAM algorithms. Recently, several learned keypoint methods have demonstrated compelling performance on challenging benchmarks. However, generating consistent and accurate training data for interest-point detection in natural images still remains challenging, especially for human annotators. We introduce IO-Net (i.e. InlierOutlierNet), a novel proxy task for the self-supervision of keypoint detection, description and matching. By making the sampling of inlier-outlier sets from point-pair correspondences fully differentiable within the keypoint learning framework, we show that are able to simultaneously self-supervise keypoint description and improve keypoint matching. Second, we introduce KeyPointNet, a keypoint-network architecture that is especially amenable to robust keypoint detection and description. We design the network to allow local keypoint aggregation to avoid artifacts due to spatial discretizations commonly used for this task, and we improve fine-grained keypoint descriptor performance by taking advantage of efficient sub-pixel convolutions to upsample the descriptor feature-maps to a higher operating resolution. Through extensive experiments and ablative analysis, we show that the proposed self-supervised keypoint learning method greatly improves the quality of feature matching and homography estimation on challenging benchmarks over the state-of-the-art.

Jiexiong Tang, Rares Ambrus, Vitor Guizilini et al.

Detecting and matching robust viewpoint-invariant keypoints is critical for visual SLAM and Structure-from-Motion. State-of-the-art learning-based methods generate training samples via homography adaptation to create 2D synthetic views with known keypoint matches from a single image. This approach, however, does not generalize to non-planar 3D scenes with illumination variations commonly seen in real-world videos. In this work, we propose self-supervised learning of depth-aware keypoints directly from unlabeled videos. We jointly learn keypoint and depth estimation networks by combining appearance and geometric matching via a differentiable structure-from-motion module based on Procrustean residual pose correction. We describe how our self-supervised keypoints can be integrated into state-of-the-art visual odometry frameworks for robust and accurate ego-motion estimation of autonomous vehicles in real-world conditions.

Vitor Guizilini, Jie Li, Rares Ambrus et al.

Dense depth estimation from a single image is a key problem in computer vision, with exciting applications in a multitude of robotic tasks. Initially viewed as a direct regression problem, requiring annotated labels as supervision at training time, in the past few years a substantial amount of work has been done in self-supervised depth training based on strong geometric cues, both from stereo cameras and more recently from monocular video sequences. In this paper we investigate how these two approaches (supervised & self-supervised) can be effectively combined, so that a depth model can learn to encode true scale from sparse supervision while achieving high fidelity local accuracy by leveraging geometric cues. To this end, we propose a novel supervised loss term that complements the widely used photometric loss, and show how it can be used to train robust semi-supervised monocular depth estimation models. Furthermore, we evaluate how much supervision is actually necessary to train accurate scale-aware monocular depth models, showing that with our proposed framework, very sparse LiDAR information, with as few as 4 beams (less than 100 valid depth values per image), is enough to achieve results competitive with the current state-of-the-art.

Rares Ambrus, Vitor Guizilini, Jie Li et al.

Learning depth and camera ego-motion from raw unlabeled RGB video streams is seeing exciting progress through self-supervision from strong geometric cues. To leverage not only appearance but also scene geometry, we propose a novel self-supervised two-stream network using RGB and inferred depth information for accurate visual odometry. In addition, we introduce a sparsity-inducing data augmentation policy for ego-motion learning that effectively regularizes the pose network to enable stronger generalization performance. As a result, we show that our proposed two-stream pose network achieves state-of-the-art results among learning-based methods on the KITTI odometry benchmark, and is especially suited for self-supervision at scale. Our experiments on a large-scale urban driving dataset of 1 million frames indicate that the performance of our proposed architecture does indeed scale progressively with more data.