Large models have shown generalization across datasets for many low-level vision tasks, like depth estimation, but no such general models exist for scene flow. Even though scene flow has wide potential use, it is not used in practice because current predictive models do not generalize well. We identify three key challenges and propose solutions for each. First, we create a method that jointly estimates geometry and motion for accurate prediction. Second, we alleviate scene flow data scarcity with a data recipe that affords us 1M annotated training samples across diverse synthetic scenes. Third, we evaluate different parameterizations for scene flow prediction and adopt a natural and effective parameterization. Our resulting model outperforms existing methods as well as baselines built on large-scale models in terms of 3D end-point error, and shows zero-shot generalization to the casually captured videos from DAVIS and the robotic manipulation scenes from RoboTAP. Overall, our approach makes scene flow prediction more practical in-the-wild.
Current image tokenization methods require a large number of tokens to capture the information contained within images. Although the amount of information varies across images, most image tokenizers only support fixed-length tokenization, leading to inefficiency in token allocation. In this study, we introduce One-D-Piece, a discrete image tokenizer designed for variable-length tokenization, achieving quality-controllable mechanism. To enable variable compression rate, we introduce a simple but effective regularization mechanism named "Tail Token Drop" into discrete one-dimensional image tokenizers. This method encourages critical information to concentrate at the head of the token sequence, enabling support of variadic tokenization, while preserving state-of-the-art reconstruction quality. We evaluate our tokenizer across multiple reconstruction quality metrics and find that it delivers significantly better perceptual quality than existing quality-controllable compression methods, including JPEG and WebP, at smaller byte sizes. Furthermore, we assess our tokenizer on various downstream computer vision tasks, including image classification, object detection, semantic segmentation, and depth estimation, confirming its adaptability to numerous applications compared to other variable-rate methods. Our approach demonstrates the versatility of variable-length discrete image tokenization, establishing a new paradigm in both compression efficiency and reconstruction performance. Finally, we validate the effectiveness of tail token drop via detailed analysis of tokenizers.
AI-driven pulse thermography (PT) has become a crucial tool in non-destructive testing (NDT), enabling automatic detection of hidden anomalies in various industrial components. Current state-of-the-art techniques feed segmentation and depth estimation networks compressed PT sequences using either Principal Component Analysis (PCA) or Thermographic Signal Reconstruction (TSR). However, treating these two modalities independently constrains the performance of PT inspection models as these representations possess complementary semantic features. To address this limitation, this work proposes PT-Fusion, a multi-modal attention-based fusion network that fuses both PCA and TSR modalities for defect segmentation and depth estimation of subsurface defects in PT setups. PT-Fusion introduces novel feature fusion modules, Encoder Attention Fusion Gate (EAFG) and Attention Enhanced Decoding Block (AEDB), to fuse PCA and TSR features for enhanced segmentation and depth estimation of subsurface defects. In addition, a novel data augmentation technique is proposed based on random data sampling from thermographic sequences to alleviate the scarcity of PT datasets. The proposed method is benchmarked against state-of-the-art PT inspection models, including U-Net, attention U-Net, and 3D-CNN on the Universit\'e Laval IRT-PVC dataset. The results demonstrate that PT-Fusion outperforms the aforementioned models in defect segmentation and depth estimation accuracies with a margin of 10%.
Tremendous progress has been made in deep stereo matching to excel on benchmark datasets through per-domain fine-tuning. However, achieving strong zero-shot generalization - a hallmark of foundation models in other computer vision tasks - remains challenging for stereo matching. We introduce FoundationStereo, a foundation model for stereo depth estimation designed to achieve strong zero-shot generalization. To this end, we first construct a large-scale (1M stereo pairs) synthetic training dataset featuring large diversity and high photorealism, followed by an automatic self-curation pipeline to remove ambiguous samples. We then design a number of network architecture components to enhance scalability, including a side-tuning feature backbone that adapts rich monocular priors from vision foundation models to mitigate the sim-to-real gap, and long-range context reasoning for effective cost volume filtering. Together, these components lead to strong robustness and accuracy across domains, establishing a new standard in zero-shot stereo depth estimation. Project page: https://nvlabs.github.io/FoundationStereo/
Stereo matching is a key technique for metric depth estimation in computer vision and robotics. Real-world challenges like occlusion and non-texture hinder accurate disparity estimation from binocular matching cues. Recently, monocular relative depth estimation has shown remarkable generalization using vision foundation models. Thus, to facilitate robust stereo matching with monocular depth cues, we incorporate a robust monocular relative depth model into the recurrent stereo-matching framework, building a new framework for depth foundation model-based stereo-matching, DEFOM-Stereo. In the feature extraction stage, we construct the combined context and matching feature encoder by integrating features from conventional CNNs and DEFOM. In the update stage, we use the depth predicted by DEFOM to initialize the recurrent disparity and introduce a scale update module to refine the disparity at the correct scale. DEFOM-Stereo is verified to have comparable performance on the Scene Flow dataset with state-of-the-art (SOTA) methods and notably shows much stronger zero-shot generalization. Moreover, DEFOM-Stereo achieves SOTA performance on the KITTI 2012, KITTI 2015, Middlebury, and ETH3D benchmarks, ranking 1st on many metrics. In the joint evaluation under the robust vision challenge, our model simultaneously outperforms previous models on the individual benchmarks. Both results demonstrate the outstanding capabilities of the proposed model.
State-of-the-art supervised stereo matching methods have achieved amazing results on various benchmarks. However, these data-driven methods suffer from generalization to real-world scenarios due to the lack of real-world annotated data. In this paper, we propose StereoGen, a novel pipeline for high-quality stereo image generation. This pipeline utilizes arbitrary single images as left images and pseudo disparities generated by a monocular depth estimation model to synthesize high-quality corresponding right images. Unlike previous methods that fill the occluded area in warped right images using random backgrounds or using convolutions to take nearby pixels selectively, we fine-tune a diffusion inpainting model to recover the background. Images generated by our model possess better details and undamaged semantic structures. Besides, we propose Training-free Confidence Generation and Adaptive Disparity Selection. The former suppresses the negative effect of harmful pseudo ground truth during stereo training, while the latter helps generate a wider disparity distribution and better synthetic images. Experiments show that models trained under our pipeline achieve state-of-the-art zero-shot generalization results among all published methods. The code will be available upon publication of the paper.
Stereo matching recovers depth from image correspondences. Existing methods struggle to handle ill-posed regions with limited matching cues, such as occlusions and textureless areas. To address this, we propose MonSter, a novel method that leverages the complementary strengths of monocular depth estimation and stereo matching. MonSter integrates monocular depth and stereo matching into a dual-branch architecture to iteratively improve each other. Confidence-based guidance adaptively selects reliable stereo cues for monodepth scale-shift recovery. The refined monodepth is in turn guides stereo effectively at ill-posed regions. Such iterative mutual enhancement enables MonSter to evolve monodepth priors from coarse object-level structures to pixel-level geometry, fully unlocking the potential of stereo matching. As shown in Fig.1, MonSter ranks 1st across five most commonly used leaderboards -- SceneFlow, KITTI 2012, KITTI 2015, Middlebury, and ETH3D. Achieving up to 49.5% improvements (Bad 1.0 on ETH3D) over the previous best method. Comprehensive analysis verifies the effectiveness of MonSter in ill-posed regions. In terms of zero-shot generalization, MonSter significantly and consistently outperforms state-of-the-art across the board. The code is publicly available at: https://github.com/Junda24/MonSter.
Birds Eye View perception models require extensive data to perform and generalize effectively. While traditional datasets often provide abundant driving scenes from diverse locations, this is not always the case. It is crucial to maximize the utility of the available training data. With the advent of large foundation models such as DINOv2 and Metric3Dv2, a pertinent question arises: can these models be integrated into existing model architectures to not only reduce the required training data but surpass the performance of current models? We choose two model architectures in the vehicle segmentation domain to alter: Lift-Splat-Shoot, and Simple-BEV. For Lift-Splat-Shoot, we explore the implementation of frozen DINOv2 for feature extraction and Metric3Dv2 for depth estimation, where we greatly exceed the baseline results by 7.4 IoU while utilizing only half the training data and iterations. Furthermore, we introduce an innovative application of Metric3Dv2's depth information as a PseudoLiDAR point cloud incorporated into the Simple-BEV architecture, replacing traditional LiDAR. This integration results in a +3 IoU improvement compared to the Camera-only model.
Recent advances in monocular depth estimation methods (MDE) and their improved accuracy open new possibilities for their applications. In this paper, we investigate how monocular depth estimates can be used for relative pose estimation. In particular, we are interested in answering the question whether using MDEs improves results over traditional point-based methods. We propose a novel framework for estimating the relative pose of two cameras from point correspondences with associated monocular depths. Since depth predictions are typically defined up to an unknown scale or even both unknown scale and shift parameters, our solvers jointly estimate the scale or both the scale and shift parameters along with the relative pose. We derive efficient solvers considering different types of depths for three camera configurations: (1) two calibrated cameras, (2) two cameras with an unknown shared focal length, and (3) two cameras with unknown different focal lengths. Our new solvers outperform state-of-the-art depth-aware solvers in terms of speed and accuracy. In extensive real experiments on multiple datasets and with various MDEs, we discuss which depth-aware solvers are preferable in which situation. The code will be made publicly available.
While recent foundation models have enabled significant breakthroughs in monocular depth estimation, a clear path towards safe and reliable deployment in the real-world remains elusive. Metric depth estimation, which involves predicting absolute distances, poses particular challenges, as even the most advanced foundation models remain prone to critical errors. Since quantifying the uncertainty has emerged as a promising endeavor to address these limitations and enable trustworthy deployment, we fuse five different uncertainty quantification methods with the current state-of-the-art DepthAnythingV2 foundation model. To cover a wide range of metric depth domains, we evaluate their performance on four diverse datasets. Our findings identify fine-tuning with the Gaussian Negative Log-Likelihood Loss (GNLL) as a particularly promising approach, offering reliable uncertainty estimates while maintaining predictive performance and computational efficiency on par with the baseline, encompassing both training and inference time. By fusing uncertainty quantification and foundation models within the context of monocular depth estimation, this paper lays a critical foundation for future research aimed at improving not only model performance but also its explainability. Extending this critical synthesis of uncertainty quantification and foundation models into other crucial tasks, such as semantic segmentation and pose estimation, presents exciting opportunities for safer and more reliable machine vision systems.
We present a novel approach for depth estimation from images captured by structured light systems. Unlike many previous methods that rely on image matching process, our approach uses a density voxel grid to represent scene geometry, which is trained via self-supervised differentiable volume rendering. Our method leverages color fields derived from projected patterns in structured light systems during the rendering process, enabling the isolated optimization of the geometry field. This contributes to faster convergence and high-quality output. Additionally, we incorporate normalized device coordinates (NDC), a distortion loss, and a novel surface-based color loss to enhance geometric fidelity. Experimental results demonstrate that our method outperforms existing matching-based techniques in geometric performance for few-shot scenarios, achieving approximately a 60% reduction in average estimated depth errors on synthetic scenes and about 30% on real-world captured scenes. Furthermore, our approach delivers fast training, with a speed roughly three times faster than previous matching-free methods that employ implicit representations.
We investigate a human-like interpretable model of video understanding. Humans recognise complex activities in video by recognising critical spatio-temporal relations among explicitly recognised objects and parts, for example, an object entering the aperture of a container. To mimic this we build on a model which uses positions of objects and hands, and their motions, to recognise the activity taking place. To improve this model we focussed on three of the most confused classes (for this model) and identified that the lack of 3D information was the major problem. To address this we extended our basic model by adding 3D awareness in two ways: (1) A state-of-the-art object detection model was fine-tuned to determine the difference between "Container" and "NotContainer" in order to integrate object shape information into the existing object features. (2) A state-of-the-art depth estimation model was used to extract depth values for individual objects and calculate depth relations to expand the existing relations used our interpretable model. These 3D extensions to our basic model were evaluated on a subset of three superficially similar "Putting" actions from the Something-Something-v2 dataset. The results showed that the container detector did not improve performance, but the addition of depth relations made a significant improvement to performance.
Monocular depth estimation (MDE) models have undergone significant advancements over recent years. Many MDE models aim to predict affine-invariant relative depth from monocular images, while recent developments in large-scale training and vision foundation models enable reasonable estimation of metric (absolute) depth. However, effectively leveraging these predictions for geometric vision tasks, in particular relative pose estimation, remains relatively under explored. While depths provide rich constraints for cross-view image alignment, the intrinsic noise and ambiguity from the monocular depth priors present practical challenges to improving upon classic keypoint-based solutions. In this paper, we develop three solvers for relative pose estimation that explicitly account for independent affine (scale and shift) ambiguities, covering both calibrated and uncalibrated conditions. We further propose a hybrid estimation pipeline that combines our proposed solvers with classic point-based solvers and epipolar constraints. We find that the affine correction modeling is beneficial to not only the relative depth priors but also, surprisingly, the ``metric" ones. Results across multiple datasets demonstrate large improvements of our approach over classic keypoint-based baselines and PnP-based solutions, under both calibrated and uncalibrated setups. We also show that our method improves consistently with different feature matchers and MDE models, and can further benefit from very recent advances on both modules. Code is available at https://github.com/MarkYu98/madpose.
The shape of human brain is complex and highly variable, with interactions between brain size, cortical folding, and age well-documented in the literature. However, few studies have explored how global brain size influences geometric features of the cortical surface derived from anatomical MRI. In this work, we focus on sulcal depth, an imaging phenotype that has gained significant attention in both basic research and clinical applications. We make key contributions to the field by: 1) providing the first quantitative analysis of how brain size affects sulcal depth measurements; 2) introducing a novel, scale-invariant method for sulcal depth estimation based on an original formalization of the problem; 3) presenting a validation framework and sharing our code and benchmark data with the community; and 4) demonstrating the biological relevance of our new sulcal depth measure using a large sample of 1,987 subjects spanning the developmental period from 26 weeks post-conception to adulthood.
Depth estimation (DE) provides spatial information about a scene and enables tasks such as 3D reconstruction, object detection, and scene understanding. Recently, there has been an increasing interest in using deep learning (DL)-based methods for DE. Traditional techniques rely on handcrafted features that often struggle to generalise to diverse scenes and require extensive manual tuning. However, DL models for DE can automatically extract relevant features from input data, adapt to various scene conditions, and generalise well to unseen environments. Numerous DL-based methods have been developed, making it necessary to survey and synthesize the state-of-the-art (SOTA). Previous reviews on DE have mainly focused on either monocular or stereo-based techniques, rather than comprehensively reviewing DE. Furthermore, to the best of our knowledge, there is no systematic literature review (SLR) that comprehensively focuses on DE. Therefore, this SLR study is being conducted. Initially, electronic databases were searched for relevant publications, resulting in 1284 publications. Using defined exclusion and quality criteria, 128 publications were shortlisted and further filtered to select 59 high-quality primary studies. These studies were analysed to extract data and answer defined research questions. Based on the results, DL methods were developed for mainly three different types of DE: monocular, stereo, and multi-view. 20 publicly available datasets were used to train, test, and evaluate DL models for DE, with KITTI, NYU Depth V2, and Make 3D being the most used datasets. 29 evaluation metrics were used to assess the performance of DE. 35 base models were reported in the primary studies, and the top five most-used base models were ResNet-50, ResNet-18, ResNet-101, U-Net, and VGG-16. Finally, the lack of ground truth data was among the most significant challenges reported by primary studies.
Safety-critical driving data is crucial for developing safe and trustworthy self-driving algorithms. Due to the scarcity of safety-critical data in naturalistic datasets, current approaches primarily utilize simulated or artificially generated images. However, there remains a gap in authenticity between these generated images and naturalistic ones. We propose a novel framework to augment the safety-critical driving data from the naturalistic dataset to address this issue. In this framework, we first detect vehicles using YOLOv5, followed by depth estimation and 3D transformation to simulate vehicle proximity and critical driving scenarios better. This allows for targeted modification of vehicle dynamics data to reflect potentially hazardous situations. Compared to the simulated or artificially generated data, our augmentation methods can generate safety-critical driving data with minimal compromise on image authenticity. Experiments using KITTI datasets demonstrate that a downstream self-driving algorithm trained on this augmented dataset performs superiorly compared to the baselines, which include SMOGN and importance sampling.
Monocular depth estimation within the diffusion-denoising paradigm demonstrates impressive generalization ability but suffers from low inference speed. Recent methods adopt a single-step deterministic paradigm to improve inference efficiency while maintaining comparable performance. However, they overlook the gap between generative and discriminative features, leading to suboptimal results. In this work, we propose DepthMaster, a single-step diffusion model designed to adapt generative features for the discriminative depth estimation task. First, to mitigate overfitting to texture details introduced by generative features, we propose a Feature Alignment module, which incorporates high-quality semantic features to enhance the denoising network's representation capability. Second, to address the lack of fine-grained details in the single-step deterministic framework, we propose a Fourier Enhancement module to adaptively balance low-frequency structure and high-frequency details. We adopt a two-stage training strategy to fully leverage the potential of the two modules. In the first stage, we focus on learning the global scene structure with the Feature Alignment module, while in the second stage, we exploit the Fourier Enhancement module to improve the visual quality. Through these efforts, our model achieves state-of-the-art performance in terms of generalization and detail preservation, outperforming other diffusion-based methods across various datasets. Our project page can be found at https://indu1ge.github.io/DepthMaster_page.
While recent depth foundation models exhibit strong zero-shot generalization, achieving accurate metric depth across diverse camera types-particularly those with large fields of view (FoV) such as fisheye and 360-degree cameras-remains a significant challenge. This paper presents Depth Any Camera (DAC), a powerful zero-shot metric depth estimation framework that extends a perspective-trained model to effectively handle cameras with varying FoVs. The framework is designed to ensure that all existing 3D data can be leveraged, regardless of the specific camera types used in new applications. Remarkably, DAC is trained exclusively on perspective images but generalizes seamlessly to fisheye and 360-degree cameras without the need for specialized training data. DAC employs Equi-Rectangular Projection (ERP) as a unified image representation, enabling consistent processing of images with diverse FoVs. Its core components include pitch-aware Image-to-ERP conversion with efficient online augmentation to simulate distorted ERP patches from undistorted inputs, FoV alignment operations to enable effective training across a wide range of FoVs, and multi-resolution data augmentation to further address resolution disparities between training and testing. DAC achieves state-of-the-art zero-shot metric depth estimation, improving $\delta_1$ accuracy by up to 50% on multiple fisheye and 360-degree datasets compared to prior metric depth foundation models, demonstrating robust generalization across camera types.
Cancer remains a significant health challenge worldwide, with a new diagnosis occurring every two minutes in the UK. Surgery is one of the main treatment options for cancer. However, surgeons rely on the sense of touch and naked eye with limited use of pre-operative image data to directly guide the excision of cancerous tissues and metastases due to the lack of reliable intraoperative visualisation tools. This leads to increased costs and harm to the patient where the cancer is removed with positive margins, or where other critical structures are unintentionally impacted. There is therefore a pressing need for more reliable and accurate intraoperative visualisation tools for minimally invasive surgery to improve surgical outcomes and enhance patient care. A recent miniaturised cancer detection probe (i.e., SENSEI developed by Lightpoint Medical Ltd.) leverages the cancer-targeting ability of nuclear agents to more accurately identify cancer intra-operatively using the emitted gamma signal. However, the use of this probe presents a visualisation challenge as the probe is non-imaging and is air-gapped from the tissue, making it challenging for the surgeon to locate the probe-sensing area on the tissue surface. Geometrically, the sensing area is defined as the intersection point between the gamma probe axis and the tissue surface in 3D space but projected onto the 2D laparoscopic image. Hence, in this thesis, tool tracking, pose estimation, and segmentation tools were developed first, followed by laparoscope image depth estimation algorithms and 3D reconstruction methods.
Accurate depth estimation is crucial for many fields, including robotics, navigation, and medical imaging. However, conventional depth sensors often produce low-resolution (LR) depth maps, making detailed scene perception challenging. To address this, enhancing LR depth maps to high-resolution (HR) ones has become essential, guided by HR-structured inputs like RGB or grayscale images. We propose a novel sensor fusion methodology for guided depth super-resolution (GDSR), a technique that combines LR depth maps with HR images to estimate detailed HR depth maps. Our key contribution is the Incremental guided attention fusion (IGAF) module, which effectively learns to fuse features from RGB images and LR depth maps, producing accurate HR depth maps. Using IGAF, we build a robust super-resolution model and evaluate it on multiple benchmark datasets. Our model achieves state-of-the-art results compared to all baseline models on the NYU v2 dataset for $\times 4$, $\times 8$, and $\times 16$ upsampling. It also outperforms all baselines in a zero-shot setting on the Middlebury, Lu, and RGB-D-D datasets. Code, environments, and models are available on GitHub.