The video action segmentation task is regularly explored under weaker forms of supervision, such as transcript supervision, where a list of actions is easier to obtain than dense frame-wise labels. In this formulation, the task presents various challenges for sequence modeling approaches due to the emphasis on action transition points, long sequence lengths, and frame contextualization, making the task well-posed for transformers. Given developments enabling transformers to scale linearly, we demonstrate through our architecture how they can be applied to improve action alignment accuracy over the equivalent RNN-based models with the attention mechanism focusing around salient action transition regions. Additionally, given the recent focus on inference-time transcript selection, we propose a supplemental transcript embedding approach to select transcripts more quickly at inference-time. Furthermore, we subsequently demonstrate how this approach can also improve the overall segmentation performance. Finally, we evaluate our proposed methods across the benchmark datasets to better understand the applicability of transformers and the importance of transcript selection on this video-driven weakly-supervised task.
As 3D object detection on point clouds relies on the geometrical relationships between the points, non-standard object shapes can hinder a method's detection capability. However, in safety-critical settings, robustness on out-of-distribution and long-tail samples is fundamental to circumvent dangerous issues, such as the misdetection of damaged or rare cars. In this work, we substantially improve the generalization of 3D object detectors to out-of-domain data by taking into account deformed point clouds during training. We achieve this with 3D-VField: a novel method that plausibly deforms objects via vectors learned in an adversarial fashion. Our approach constrains 3D points to slide along their sensor view rays while neither adding nor removing any of them. The obtained vectors are transferrable, sample-independent and preserve shape smoothness and occlusions. By augmenting normal samples with the deformations produced by these vector fields during training, we significantly improve robustness against differently shaped objects, such as damaged/deformed cars, even while training only on KITTI. Towards this end, we propose and share open source CrashD: a synthetic dataset of realistic damaged and rare cars, with a variety of crash scenarios. Extensive experiments on KITTI, Waymo, our CrashD and SUN RGB-D show the high generalizability of our techniques to out-of-domain data, different models and sensors, namely LiDAR and ToF cameras, for both indoor and outdoor scenes. Our CrashD dataset is available at https://crashd-cars.github.io.
Recently Implicit Neural Representations (INRs) gained attention as a novel and effective representation for various data types. Thus far, prior work mostly focused on optimizing their reconstruction performance. This work investigates INRs from a novel perspective, i.e., as a tool for image compression. To this end, we propose the first comprehensive compression pipeline based on INRs including quantization, quantization-aware retraining and entropy coding. Encoding with INRs, i.e. overfitting to a data sample, is typically orders of magnitude slower. To mitigate this drawback, we leverage meta-learned initializations based on MAML to reach the encoding in fewer gradient updates which also generally improves rate-distortion performance of INRs. We find that our approach to source compression with INRs vastly outperforms similar prior work, is competitive with common compression algorithms designed specifically for images and closes the gap to state-of-the-art learned approaches based on Rate-Distortion Autoencoders. Moreover, we provide an extensive ablation study on the importance of individual components of our method which we hope facilitates future research on this novel approach to image compression.
With the advent of deep learning, estimating depth from a single RGB image has recently received a lot of attention, being capable of empowering many different applications ranging from path planning for robotics to computational cinematography. Nevertheless, while the depth maps are in their entirety fairly reliable, the estimates around object discontinuities are still far from satisfactory. This can be contributed to the fact that the convolutional operator naturally aggregates features across object discontinuities, resulting in smooth transitions rather than clear boundaries. Therefore, in order to circumvent this issue, we propose a novel convolutional operator which is explicitly tailored to avoid feature aggregation of different object parts. In particular, our method is based on estimating per-part depth values by means of superpixels. The proposed convolutional operator, which we dub "Instance Convolution", then only considers each object part individually on the basis of the estimated superpixels. Our evaluation with respect to the NYUv2 as well as the iBims dataset clearly demonstrates the superiority of Instance Convolutions over the classical convolution at estimating depth around occlusion boundaries, while producing comparable results elsewhere. Code will be made publicly available upon acceptance.
Parts represent a basic unit of geometric and semantic similarity across different objects. We argue that part knowledge should be composable beyond the observed object classes. Towards this, we present 3D Compositional Zero-shot Learning as a problem of part generalization from seen to unseen object classes for semantic segmentation. We provide a structured study through benchmarking the task with the proposed Compositional-PartNet dataset. This dataset is created by processing the original PartNet to maximize part overlap across different objects. The existing point cloud part segmentation methods fail to generalize to unseen object classes in this setting. As a solution, we propose DeCompositional Consensus, which combines a part segmentation network with a part scoring network. The key intuition to our approach is that a segmentation mask over some parts should have a consensus with its part scores when each part is taken apart. The two networks reason over different part combinations defined in a per-object part prior to generate the most suitable segmentation mask. We demonstrate that our method allows compositional zero-shot segmentation and generalized zero-shot classification, and establishes the state of the art on both tasks.
Recent advances in machine learning have created increasing interest in solving visual computing problems using a class of coordinate-based neural networks that parametrize physical properties of scenes or objects across space and time. These methods, which we call neural fields, have seen successful application in the synthesis of 3D shapes and image, animation of human bodies, 3D reconstruction, and pose estimation. However, due to rapid progress in a short time, many papers exist but a comprehensive review and formulation of the problem has not yet emerged. In this report, we address this limitation by providing context, mathematical grounding, and an extensive review of literature on neural fields. This report covers research along two dimensions. In Part I, we focus on techniques in neural fields by identifying common components of neural field methods, including different representations, architectures, forward mapping, and generalization methods. In Part II, we focus on applications of neural fields to different problems in visual computing, and beyond (e.g., robotics, audio). Our review shows the breadth of topics already covered in visual computing, both historically and in current incarnations, demonstrating the improved quality, flexibility, and capability brought by neural fields methods. Finally, we present a companion website that contributes a living version of this review that can be continually updated by the community.
Accurate and reliable localization is a fundamental requirement for autonomous vehicles to use map information in higher-level tasks such as navigation or planning. In this paper, we present a novel approach to vehicle localization in dense semantic maps, including vectorized high-definition maps or 3D meshes, using semantic segmentation from a monocular camera. We formulate the localization task as a direct image alignment problem on semantic images, which allows our approach to robustly track the vehicle pose in semantically labeled maps by aligning virtual camera views rendered from the map to sequences of semantically segmented camera images. In contrast to existing visual localization approaches, the system does not require additional keypoint features, handcrafted localization landmark extractors or expensive LiDAR sensors. We demonstrate the wide applicability of our method on a diverse set of semantic mesh maps generated from stereo or LiDAR as well as manually annotated HD maps and show that it achieves reliable and accurate localization in real-time.
Estimating the uncertainty of a neural network plays a fundamental role in safety-critical settings. In perception for autonomous driving, measuring the uncertainty means providing additional calibrated information to downstream tasks, such as path planning, that can use it towards safe navigation. In this work, we propose a novel sampling-free uncertainty estimation method for object detection. We call it CertainNet, and it is the first to provide separate uncertainties for each output signal: objectness, class, location and size. To achieve this, we propose an uncertainty-aware heatmap, and exploit the neighboring bounding boxes provided by the detector at inference time. We evaluate the detection performance and the quality of the different uncertainty estimates separately, also with challenging out-of-domain samples: BDD100K and nuImages with models trained on KITTI. Additionally, we propose a new metric to evaluate location and size uncertainties. When transferring to unseen datasets, CertainNet generalizes substantially better than previous methods and an ensemble, while being real-time and providing high quality and comprehensive uncertainty estimates.
In this paper, a method for dense semantic 3D scene reconstruction from an RGB-D sequence is proposed to solve high-level scene understanding tasks. First, each RGB-D pair is consistently segmented into 2D semantic maps based on a camera tracking backbone that propagates objects' labels with high probabilities from full scans to corresponding ones of partial views. Then a dense 3D mesh model of an unknown environment is incrementally generated from the input RGB-D sequence. Benefiting from 2D consistent semantic segments and the 3D model, a novel semantic projection block (SP-Block) is proposed to extract deep feature volumes from 2D segments of different views. Moreover, the semantic volumes are fused into deep volumes from a point cloud encoder to make the final semantic segmentation. Extensive experimental evaluations on public datasets show that our system achieves accurate 3D dense reconstruction and state-of-the-art semantic prediction performances simultaneously.
Depth estimation from monocular images is an important task in localization and 3D reconstruction pipelines for bronchoscopic navigation. Various supervised and self-supervised deep learning-based approaches have proven themselves on this task for natural images. However, the lack of labeled data and the bronchial tissue's feature-scarce texture make the utilization of these methods ineffective on bronchoscopic scenes. In this work, we propose an alternative domain-adaptive approach. Our novel two-step structure first trains a depth estimation network with labeled synthetic images in a supervised manner; then adopts an unsupervised adversarial domain feature adaptation scheme to improve the performance on real images. The results of our experiments show that the proposed method improves the network's performance on real images by a considerable margin and can be employed in 3D reconstruction pipelines.