While recent model-free Reinforcement Learning (RL) methods have demonstrated human-level effectiveness in gaming environments, their success in everyday tasks like visual navigation has been limited, particularly under significant appearance variations. This limitation arises from (i) poor sample efficiency and (ii) over-fitting to training scenarios. To address these challenges, we present a world model that learns invariant features using (i) contrastive unsupervised learning and (ii) an intervention-invariant regularizer. Learning an explicit representation of the world dynamics i.e. a world model, improves sample efficiency while contrastive learning implicitly enforces learning of invariant features, which improves generalization. However, the na\"ive integration of contrastive loss to world models is not good enough, as world-model-based RL methods independently optimize representation learning and agent policy. To overcome this issue, we propose an intervention-invariant regularizer in the form of an auxiliary task such as depth prediction, image denoising, image segmentation, etc., that explicitly enforces invariance to style interventions. Our method outperforms current state-of-the-art model-based and model-free RL methods and significantly improves on out-of-distribution point navigation tasks evaluated on the iGibson benchmark. With only visual observations, we further demonstrate that our approach outperforms recent language-guided foundation models for point navigation, which is essential for deployment on robots with limited computation capabilities. Finally, we demonstrate that our proposed model excels at the sim-to-real transfer of its perception module on the Gibson benchmark.
Camera-based bird-eye-view (BEV) perception paradigm has made significant progress in the autonomous driving field. Under such a paradigm, accurate BEV representation construction relies on reliable depth estimation for multi-camera images. However, existing approaches exhaustively predict depths for every pixel without prioritizing objects, which are precisely the entities requiring detection in the 3D space. To this end, we propose IA-BEV, which integrates image-plane instance awareness into the depth estimation process within a BEV-based detector. First, a category-specific structural priors mining approach is proposed for enhancing the efficacy of monocular depth generation. Besides, a self-boosting learning strategy is further proposed to encourage the model to place more emphasis on challenging objects in computation-expensive temporal stereo matching. Together they provide advanced depth estimation results for high-quality BEV features construction, benefiting the ultimate 3D detection. The proposed method achieves state-of-the-art performances on the challenging nuScenes benchmark, and extensive experimental results demonstrate the effectiveness of our designs.
Learning-based monocular depth estimation leverages geometric priors present in the training data to enable metric depth perception from a single image, a traditionally ill-posed problem. However, these priors are often specific to a particular domain, leading to limited generalization performance on unseen data. Apart from the well studied environmental domain gap, monocular depth estimation is also sensitive to the domain gap induced by varying camera parameters, an aspect that is often overlooked in current state-of-the-art approaches. This issue is particularly evident in autonomous driving scenarios, where datasets are typically collected with a single vehicle-camera setup, leading to a bias in the training data due to a fixed perspective geometry. In this paper, we challenge this trend and introduce GenDepth, a novel model capable of performing metric depth estimation for arbitrary vehicle-camera setups. To address the lack of data with sufficiently diverse camera parameters, we first create a bespoke synthetic dataset collected with different vehicle-camera systems. Then, we design GenDepth to simultaneously optimize two objectives: (i) equivariance to the camera parameter variations on synthetic data, (ii) transferring the learned equivariance to real-world environmental features using a single real-world dataset with a fixed vehicle-camera system. To achieve this, we propose a novel embedding of camera parameters as the ground plane depth and present a novel architecture that integrates these embeddings with adversarial domain alignment. We validate GenDepth on several autonomous driving datasets, demonstrating its state-of-the-art generalization capability for different vehicle-camera systems.
Semantic scene completion (SSC) aims to predict the semantic occupancy of each voxel in the entire 3D scene from limited observations, which is an emerging and critical task for autonomous driving. Recently, many studies have turned to camera-based SSC solutions due to the richer visual cues and cost-effectiveness of cameras. However, existing methods usually rely on sophisticated and heavy 3D models to process the lifted 3D features directly, which are not discriminative enough for clear segmentation boundaries. In this paper, we adopt the dense-sparse-dense design and propose a one-stage camera-based SSC framework, termed SGN, to propagate semantics from the semantic-aware seed voxels to the whole scene based on spatial geometry cues. Firstly, to exploit depth-aware context and dynamically select sparse seed voxels, we redesign the sparse voxel proposal network to process points generated by depth prediction directly with the coarse-to-fine paradigm. Furthermore, by designing hybrid guidance (sparse semantic and geometry guidance) and effective voxel aggregation for spatial geometry cues, we enhance the feature separation between different categories and expedite the convergence of semantic propagation. Finally, we devise the multi-scale semantic propagation module for flexible receptive fields while reducing the computation resources. Extensive experimental results on the SemanticKITTI and SSCBench-KITTI-360 datasets demonstrate the superiority of our SGN over existing state-of-the-art methods. And even our lightweight version SGN-L achieves notable scores of 14.80\% mIoU and 45.45\% IoU on SeamnticKITTI validation with only 12.5 M parameters and 7.16 G training memory. Code is available at https://github.com/Jieqianyu/SGN.
Objects undergo varying amounts of perspective distortion as they move across a camera's field of view. Models for predicting 3D from a single image often work with crops around the object of interest and ignore the location of the object in the camera's field of view. We note that ignoring this location information further exaggerates the inherent ambiguity in making 3D inferences from 2D images and can prevent models from even fitting to the training data. To mitigate this ambiguity, we propose Intrinsics-Aware Positional Encoding (KPE), which incorporates information about the location of crops in the image and camera intrinsics. Experiments on three popular 3D-from-a-single-image benchmarks: depth prediction on NYU, 3D object detection on KITTI & nuScenes, and predicting 3D shapes of articulated objects on ARCTIC, show the benefits of KPE.
Current deep networks are very data-hungry and benefit from training on largescale datasets, which are often time-consuming to collect and annotate. By contrast, synthetic data can be generated infinitely using generative models such as DALL-E and diffusion models, with minimal effort and cost. In this paper, we present DatasetDM, a generic dataset generation model that can produce diverse synthetic images and the corresponding high-quality perception annotations (e.g., segmentation masks, and depth). Our method builds upon the pre-trained diffusion model and extends text-guided image synthesis to perception data generation. We show that the rich latent code of the diffusion model can be effectively decoded as accurate perception annotations using a decoder module. Training the decoder only needs less than 1% (around 100 images) manually labeled images, enabling the generation of an infinitely large annotated dataset. Then these synthetic data can be used for training various perception models for downstream tasks. To showcase the power of the proposed approach, we generate datasets with rich dense pixel-wise labels for a wide range of downstream tasks, including semantic segmentation, instance segmentation, and depth estimation. Notably, it achieves 1) state-of-the-art results on semantic segmentation and instance segmentation; 2) significantly more robust on domain generalization than using the real data alone; and state-of-the-art results in zero-shot segmentation setting; and 3) flexibility for efficient application and novel task composition (e.g., image editing). The project website and code can be found at https://weijiawu.github.io/DatasetDM_page/ and https://github.com/showlab/DatasetDM, respectively
This paper introduces a novel approach to the fine alignment of images in a burst captured by a handheld camera. In contrast to traditional techniques that estimate two-dimensional transformations between frame pairs or rely on discrete correspondences, the proposed algorithm establishes dense correspondences by optimizing both the camera motion and surface depth and orientation at every pixel. This approach improves alignment, particularly in scenarios with parallax challenges. Extensive experiments with synthetic bursts featuring small and even tiny baselines demonstrate that it outperforms the best optical flow methods available today in this setting, without requiring any training. Beyond enhanced alignment, our method opens avenues for tasks beyond simple image restoration, such as depth estimation and 3D reconstruction, as supported by promising preliminary results. This positions our approach as a versatile tool for various burst image processing applications.
3D sensing for monocular in-the-wild images, e.g., depth estimation and 3D object detection, has become increasingly important. However, the unknown intrinsic parameter hinders their development and deployment. Previous methods for the monocular camera calibration rely on specific 3D objects or strong geometry prior, such as using a checkerboard or imposing a Manhattan World assumption. This work solves the problem from the other perspective by exploiting the monocular 3D prior. Our method is assumption-free and calibrates the complete $4$ Degree-of-Freedom (DoF) intrinsic parameters. First, we demonstrate intrinsic is solved from two well-studied monocular priors, i.e., monocular depthmap, and surface normal map. However, this solution imposes a low-bias and low-variance requirement for depth estimation. Alternatively, we introduce a novel monocular 3D prior, the incidence field, defined as the incidence rays between points in 3D space and pixels in the 2D imaging plane. The incidence field is a pixel-wise parametrization of the intrinsic invariant to image cropping and resizing. With the estimated incidence field, a robust RANSAC algorithm recovers intrinsic. We demonstrate the effectiveness of our method by showing superior performance on synthetic and zero-shot testing datasets. Beyond calibration, we demonstrate downstream applications in image manipulation detection & restoration, uncalibrated two-view pose estimation, and 3D sensing. Codes, models, and data will be held in https://github.com/ShngJZ/WildCamera.
In this paper, we introduce a novel training method for making any monocular depth network learn absolute scale and estimate metric road-scene depth just from regular training data, i.e., driving videos. We refer to this training framework as FUMET. The key idea is to leverage cars found on the road as sources of scale supervision and to incorporate them in network training robustly. FUMET detects and estimates the sizes of cars in a frame and aggregates scale information extracted from them into an estimate of the camera height whose consistency across the entire video sequence is enforced as scale supervision. This realizes robust unsupervised training of any, otherwise scale-oblivious, monocular depth network so that they become not only scale-aware but also metric-accurate without the need for auxiliary sensors and extra supervision. Extensive experiments on the KITTI and the Cityscapes datasets show the effectiveness of FUMET, which achieves state-of-the-art accuracy. We also show that FUMET enables training on mixed datasets of different camera heights, which leads to larger-scale training and better generalization. Metric depth reconstruction is essential in any road-scene visual modeling, and FUMET democratizes its deployment by establishing the means to convert any model into a metric depth estimator.
In magnetic resonance imaging (MRI), slice-to-volume reconstruction (SVR) refers to computational reconstruction of an unknown 3D magnetic resonance volume from stacks of 2D slices corrupted by motion. While promising, current SVR methods require multiple slice stacks for accurate 3D reconstruction, leading to long scans and limiting their use in time-sensitive applications such as fetal fMRI. Here, we propose a SVR method that overcomes the shortcomings of previous work and produces state-of-the-art reconstructions in the presence of extreme inter-slice motion. Inspired by the recent success of single-view depth estimation methods, we formulate SVR as a single-stack motion estimation task and train a fully convolutional network to predict a motion stack for a given slice stack, producing a 3D reconstruction as a byproduct of the predicted motion. Extensive experiments on the SVR of adult and fetal brains demonstrate that our fully convolutional method is twice as accurate as previous SVR methods. Our code is available at github.com/seannz/svr.
Monocular depth estimation (MDE) is a fundamental topic of geometric computer vision and a core technique for many downstream applications. Recently, several methods reframe the MDE as a classification-regression problem where a linear combination of probabilistic distribution and bin centers is used to predict depth. In this paper, we propose a novel concept of iterative elastic bins (IEBins) for the classification-regression-based MDE. The proposed IEBins aims to search for high-quality depth by progressively optimizing the search range, which involves multiple stages and each stage performs a finer-grained depth search in the target bin on top of its previous stage. To alleviate the possible error accumulation during the iterative process, we utilize a novel elastic target bin to replace the original target bin, the width of which is adjusted elastically based on the depth uncertainty. Furthermore, we develop a dedicated framework composed of a feature extractor and an iterative optimizer that has powerful temporal context modeling capabilities benefiting from the GRU-based architecture. Extensive experiments on the KITTI, NYU-Depth-v2 and SUN RGB-D datasets demonstrate that the proposed method surpasses prior state-of-the-art competitors. The source code is publicly available at https://github.com/ShuweiShao/IEBins.
Diffusion Handles is a novel approach to enabling 3D object edits on diffusion images. We accomplish these edits using existing pre-trained diffusion models, and 2D image depth estimation, without any fine-tuning or 3D object retrieval. The edited results remain plausible, photo-real, and preserve object identity. Diffusion Handles address a critically missing facet of generative image based creative design, and significantly advance the state-of-the-art in generative image editing. Our key insight is to lift diffusion activations for an object to 3D using a proxy depth, 3D-transform the depth and associated activations, and project them back to image space. The diffusion process applied to the manipulated activations with identity control, produces plausible edited images showing complex 3D occlusion and lighting effects. We evaluate Diffusion Handles: quantitatively, on a large synthetic data benchmark; and qualitatively by a user study, showing our output to be more plausible, and better than prior art at both, 3D editing and identity control. Project Webpage: https://diffusionhandles.github.io/
Single image depth estimation is a foundational task in computer vision and generative modeling. However, prevailing depth estimation models grapple with accommodating the increasing resolutions commonplace in today's consumer cameras and devices. Existing high-resolution strategies show promise, but they often face limitations, ranging from error propagation to the loss of high-frequency details. We present PatchFusion, a novel tile-based framework with three key components to improve the current state of the art: (1) A patch-wise fusion network that fuses a globally-consistent coarse prediction with finer, inconsistent tiled predictions via high-level feature guidance, (2) A Global-to-Local (G2L) module that adds vital context to the fusion network, discarding the need for patch selection heuristics, and (3) A Consistency-Aware Training (CAT) and Inference (CAI) approach, emphasizing patch overlap consistency and thereby eradicating the necessity for post-processing. Experiments on UnrealStereo4K, MVS-Synth, and Middleburry 2014 demonstrate that our framework can generate high-resolution depth maps with intricate details. PatchFusion is independent of the base model for depth estimation. Notably, our framework built on top of SOTA ZoeDepth brings improvements for a total of 17.3% and 29.4% in terms of the root mean squared error (RMSE) on UnrealStereo4K and MVS-Synth, respectively.
Monocular depth estimation using Convolutional Neural Networks (CNNs) has shown impressive performance in outdoor driving scenes. However, self-supervised learning of indoor depth from monocular sequences is quite challenging for researchers because of the following two main reasons. One is the large areas of low-texture regions and the other is the complex ego-motion on indoor training datasets. In this work, our proposed method, named IndoorDepth, consists of two innovations. In particular, we first propose a novel photometric loss with improved structural similarity (SSIM) function to tackle the challenge from low-texture regions. Moreover, in order to further mitigate the issue of inaccurate ego-motion prediction, multiple photometric losses at different stages are used to train a deeper pose network with two residual pose blocks. Subsequent ablation study can validate the effectiveness of each new idea. Experiments on the NYUv2 benchmark demonstrate that our IndoorDepth outperforms the previous state-of-the-art methods by a large margin. In addition, we also validate the generalization ability of our method on ScanNet dataset. Code is availabe at https://github.com/fcntes/IndoorDepth.
While perspective is a well-studied topic in art, it is generally taken for granted in images. However, for the recent wave of high-quality image synthesis methods such as latent diffusion models, perspective accuracy is not an explicit requirement. Since these methods are capable of outputting a wide gamut of possible images, it is difficult for these synthesized images to adhere to the principles of linear perspective. We introduce a novel geometric constraint in the training process of generative models to enforce perspective accuracy. We show that outputs of models trained with this constraint both appear more realistic and improve performance of downstream models trained on generated images. Subjective human trials show that images generated with latent diffusion models trained with our constraint are preferred over images from the Stable Diffusion V2 model 70% of the time. SOTA monocular depth estimation models such as DPT and PixelFormer, fine-tuned on our images, outperform the original models trained on real images by up to 7.03% in RMSE and 19.3% in SqRel on the KITTI test set for zero-shot transfer.
We present a new approach, termed GPS-Gaussian, for synthesizing novel views of a character in a real-time manner. The proposed method enables 2K-resolution rendering under a sparse-view camera setting. Unlike the original Gaussian Splatting or neural implicit rendering methods that necessitate per-subject optimizations, we introduce Gaussian parameter maps defined on the source views and regress directly Gaussian Splatting properties for instant novel view synthesis without any fine-tuning or optimization. To this end, we train our Gaussian parameter regression module on a large amount of human scan data, jointly with a depth estimation module to lift 2D parameter maps to 3D space. The proposed framework is fully differentiable and experiments on several datasets demonstrate that our method outperforms state-of-the-art methods while achieving an exceeding rendering speed.
Monocular depth estimation is a fundamental computer vision task. Recovering 3D depth from a single image is geometrically ill-posed and requires scene understanding, so it is not surprising that the rise of deep learning has led to a breakthrough. The impressive progress of monocular depth estimators has mirrored the growth in model capacity, from relatively modest CNNs to large Transformer architectures. Still, monocular depth estimators tend to struggle when presented with images with unfamiliar content and layout, since their knowledge of the visual world is restricted by the data seen during training, and challenged by zero-shot generalization to new domains. This motivates us to explore whether the extensive priors captured in recent generative diffusion models can enable better, more generalizable depth estimation. We introduce Marigold, a method for affine-invariant monocular depth estimation that is derived from Stable Diffusion and retains its rich prior knowledge. The estimator can be fine-tuned in a couple of days on a single GPU using only synthetic training data. It delivers state-of-the-art performance across a wide range of datasets, including over 20% performance gains in specific cases. Project page: https://marigoldmonodepth.github.io.
We introduce Generative Infinite-Vocabulary Transformers (GIVT) which generate vector sequences with real-valued entries, instead of discrete tokens from a finite vocabulary. To this end, we propose two surprisingly simple modifications to decoder-only transformers: 1) at the input, we replace the finite-vocabulary lookup table with a linear projection of the input vectors; and 2) at the output, we replace the logits prediction (usually mapped to a categorical distribution) with the parameters of a multivariate Gaussian mixture model. Inspired by the image-generation paradigm of VQ-GAN and MaskGIT, where transformers are used to model the discrete latent sequences of a VQ-VAE, we use GIVT to model the unquantized real-valued latent sequences of a $\beta$-VAE. In class-conditional image generation GIVT outperforms VQ-GAN (and improved variants thereof) as well as MaskGIT, and achieves performance competitive with recent latent diffusion models. Finally, we obtain strong results outside of image generation when applying GIVT to panoptic segmentation and depth estimation with a VAE variant of the UViM framework.
Recently, the rise of query-based Transformer decoders is reshaping camera-based 3D object detection. These query-based decoders are surpassing the traditional dense BEV (Bird's Eye View)-based methods. However, we argue that dense BEV frameworks remain important due to their outstanding abilities in depth estimation and object localization, depicting 3D scenes accurately and comprehensively. This paper aims to address the drawbacks of the existing dense BEV-based 3D object detectors by introducing our proposed enhanced components, including a CRF-modulated depth estimation module enforcing object-level consistencies, a long-term temporal aggregation module with extended receptive fields, and a two-stage object decoder combining perspective techniques with CRF-modulated depth embedding. These enhancements lead to a "modernized" dense BEV framework dubbed BEVNeXt. On the nuScenes benchmark, BEVNeXt outperforms both BEV-based and query-based frameworks under various settings, achieving a state-of-the-art result of 64.2 NDS on the nuScenes test set. Code will be available at \url{https://github.com/woxihuanjiangguo/BEVNeXt}.
It is widely believed that sparse supervision is worse than dense supervision in the field of depth completion, but the underlying reasons for this are rarely discussed. To this end, we revisit the task of radar-camera depth completion and present a new method with sparse LiDAR supervision to outperform previous dense LiDAR supervision methods in both accuracy and speed. Specifically, when trained by sparse LiDAR supervision, depth completion models usually output depth maps containing significant stripe-like artifacts. We find that such a phenomenon is caused by the implicitly learned positional distribution pattern from sparse LiDAR supervision, termed as LiDAR Distribution Leakage (LDL) in this paper. Based on such understanding, we present a novel Disruption-Compensation radar-camera depth completion framework to address this issue. The Disruption part aims to deliberately disrupt the learning of LiDAR distribution from sparse supervision, while the Compensation part aims to leverage 3D spatial and 2D semantic information to compensate for the information loss of previous disruptions. Extensive experimental results demonstrate that by reducing the impact of LDL, our framework with sparse supervision outperforms the state-of-the-art dense supervision methods with 11.6% improvement in Mean Absolute Error (MAE)} and 1.6x speedup in Frame Per Second (FPS)}. The code is available at https://github.com/megvii-research/Sparse-Beats-Dense.