Adversarial attacks aim to perturb images such that a predictor outputs incorrect results. Due to the limited research in structured attacks, imposing consistency checks on natural multi-object scenes is a practical defense against conventional adversarial attacks. More desired attacks should be able to fool defenses with such consistency checks. Therefore, we present the first approach GLOW that copes with various attack requests by generating global layout-aware adversar-ial attacks, in which both categorical and geometric layout constraints are explicitly established. Specifically, we focus on object detection tasks and given a victim image, GLOW first localizes victim objects according to target labels. And then it generates multiple attack plans, together with their context-consistency scores. GLOW, on the one hand, is ca-pable of handling various types of requests, including single or multiple victim objects, with or without specified victim objects. On the other hand, it produces a consistency score for each attack plan, reflecting the overall contextual consistency that both semantic category and global scene layout are considered. We conduct our experiments on MS COCO and Pascal. Extensive experimental results demonstrate that we can achieve about 30% average relative improvement compared to state-of-the-art methods in conventional single object attack request; Moreover, such superiority is also valid across more generic attack requests, under both white-box and zero-query black-box settings. Finally, we conduct comprehensive human analysis, which not only validates our claim further but also provides strong evidence that our evaluation metrics reflect human reviews well.

Photorealistic simulation plays a crucial role in applications such as autonomous driving, where advances in neural radiance fields (NeRFs) may allow better scalability through the automatic creation of digital 3D assets. However, reconstruction quality suffers on street scenes due to largely collinear camera motions and sparser samplings at higher speeds. On the other hand, the application often demands rendering from camera views that deviate from the inputs to accurately simulate behaviors like lane changes. In this paper, we propose several insights that allow a better utilization of Lidar data to improve NeRF quality on street scenes. First, our framework learns a geometric scene representation from Lidar, which are fused with the implicit grid-based representation for radiance decoding, thereby supplying stronger geometric information offered by explicit point cloud. Sec-ond, we put forth a robust occlusion-aware depth supervision scheme, which allows utilizing densified Lidar points by accumulation. Third, we generate augmented training views from Lidar points for further improvement. Our insights translate to largely improved novel view synthesis under real driving scenes.

This work focuses on the active learning in video captioning. In particular, we propose to address the learnability problem in active learning, which has been brought up by collective outliers in video captioning and neglected in the literature. To start with, we conduct a comprehensive study of collective outliers, exploring their hard-to-learn property and concluding that ground truth inconsistency is one of the main causes. Motivated by this, we design a novel active learning algorithm that takes three complementary aspects, namely learnability, diversity, and uncertainty, into account. Ideally, learnability is reflected by ground truth consistency. Under the active learning scenario where ground truths are not available until human involvement, we measure the consistency on estimated ground truths, where predictions from off-the-shelf models are utilized as approximations to ground truths. These predictions are further used to estimate sample frequency and reliability, evincing the diversity and uncertainty respectively. With the help of our novel caption-wise active learning protocol, our algorithm is capable of leveraging knowledge from humans in a more effective yet intellectual manner. Results on publicly available video captioning datasets with diverse video captioning models demonstrate that our algorithm outperforms SOTA active learning methods by a large margin, e.g. we achieve about 103% of full performance on CIDEr with 25% of human annotations on MSR-VTT.

Monocular 3D object localization in driving scenes is a crucial task, but challenging due to its ill-posed nature. Estimating 3D coordinates for each pixel on the object surface holds great potential as it provides dense 2D-3D geometric constraints for the underlying PnP problem. However, high-quality ground truth supervision is not available in driving scenes due to sparsity and various artifacts of Lidar data, as well as the practical infeasibility of collecting per-instance CAD models. In this work, we present NeurOCS, a framework that uses instance masks and 3D boxes as input to learn 3D object shapes by means of differentiable rendering, which further serves as supervision for learning dense object coordinates. Our approach rests on insights in learning a category-level shape prior directly from real driving scenes, while properly handling single-view ambiguities. Furthermore, we study and make critical design choices to learn object coordinates more effectively from an object-centric view. Altogether, our framework leads to new state-of-the-art in monocular 3D localization that ranks 1st on the KITTI-Object [16] benchmark among published monocular methods.

Convolutional Neural Networks (CNNs) are vulnerable to adversarial attacks in which visually imperceptible perturbations can deceive CNN-based models. While current research on adversarial attacks in single object tracking exists, it overlooks a critical aspect of manipulating predicted trajectories to follow user-defined paths regardless of the actual location of the targeted object. To address this, we propose the very first white-box attack algorithm that is capable of deceiving victim trackers by compelling them to generate trajectories that adhere to predetermined counterfeit paths. Specifically, we focus on Siamese-based trackers as our victim models. Given an arbitrary counterfeit path, we first decompose it into discrete target locations in each frame, with the assumption of constant velocity. These locations are converted to heatmap anchors, which represent the offset of their location from the target object's location in the previous frame. Later on, we design a novel loss function to minimize the gap between above-mentioned anchors and our predicted ones. Finally, the gradients computed by such loss are used to update the original video, resulting in our adversarial video. To validate our ideas, we design three sets of counterfeit paths as well as novel evaluation metrics to measure the path-following properties. Experiments with two victim models on three publicly available datasets, OTB100, VOT2018, and VOT2016, demonstrate that our algorithm not only outperforms SOTA methods significantly under conventional evaluation metrics, e.g. 90% and 68.4% precision and successful rate drop on OTB100, but also follows the counterfeit paths well, which is beyond any existing attack methods. The source code is available at https://github.com/loushengtao/Follow-me.

Conventional video captioning methods are either stage-wise or simple end-to-end. While the former might introduce additional noise when exploiting off-the-shelf models to provide extra information, the latter suffers from lacking high-level cues. Therefore, a more desired framework should be able to capture multi-aspects of videos consistently. To this end, we present a concept-aware and task-specific model named CAT that accounts for both low-level visual and high-level concept cues, and incorporates them effectively in an end-to-end manner. Specifically, low-level visual and high-level concept features are obtained from the video transformer and concept parser of CAT. And a concept loss is further introduced to regularize the learning process of concept parser w.r.t. generated pseudo ground truth. To combine multi-level features, a caption transformer is later introduced in CAT, where visual and concept features are the inputs and caption is its output. In particular, we make critical design choices in the caption transformer to learn to exploit these cues with a multi-modal graph. This is achieved by a graph loss that enforces effective learning of intra and inter correlations between multi-level cues. Extensive experiments on three benchmark datasets demonstrate that CAT achieves 2.3 and 0.7 improvements in the CIDEr metric on MSVD and MSR-VTT compared to the state-of-the-art method SwinBERT and also achieves a competitive result on VATEX.

This paper aims to address the single image gaze target detection problem. Conventional methods either focus on 2D visual cues or exploit additional depth information in a very coarse manner. In this work, we propose to explicitly and effectively model 3D geometry under challenging scenario where only 2D annotations are available. We first obtain 3D point clouds of given scene with estimated depth and reference objects. Then we figure out the front-most points in all possible 3D directions of given person. These points are later leveraged in our ESCNet model. Specifically, ESCNet consists of geometry and scene parsing modules. The former produces an initial heatmap inferring the probability that each front-most point has been looking at according to estimated 3D gaze direction. And the latter further explores scene contextual cues to regulate detection results. We validate our idea on two publicly available dataset, GazeFollow and VideoAttentionTarget, and demon-strate the state-of-the-art performance. Our method also beats the human in terms of AUC on GazeFollow. Our code can be found here https://github.com/bjj9/ESCNet.

We propose an end-to-end network that takes a single perspective RGB image of a complex road scene as input, to produce occlusion-reasoned layouts in perspective space as well as a parametric bird's-eye-view (BEV) space. In contrast to prior works that require dense supervision such as semantic labels in perspective view, our method only requires human annotations for parametric attributes that are cheaper and less ambiguous to obtain. To solve this challenging task, our design is comprised of modules that incorporate inductive biases to learn occlusion-reasoning, geometric transformation and semantic abstraction, where each module may be supervised by appropriately transforming the parametric annotations. We demonstrate how our design choices and proposed deep supervision help achieve meaningful representations and accurate predictions. We validate our approach on two public datasets, KITTI and NuScenes, to achieve state-of-the-art results with considerably less human supervision.

Test-time adaptation approaches have recently emerged as a practical solution for handling domain shift without access to the source domain data. In this paper, we propose and explore a new multi-modal extension of test-time adaptation for 3D semantic segmentation. We find that, directly applying existing methods usually results in performance instability at test time, because multi-modal input is not considered jointly. To design a framework that can take full advantage of multi-modality, where each modality provides regularized self-supervisory signals to other modalities, we propose two complementary modules within and across the modalities. First, Intra-modal Pseudo-label Generation (Intra-PG) is introduced to obtain reliable pseudo labels within each modality by aggregating information from two models that are both pre-trained on source data but updated with target data at different paces. Second, Inter-modal Pseudo-label Refinement (Inter-PR) adaptively selects more reliable pseudo labels from different modalities based on a proposed consistency scheme. Experiments demonstrate that our regularized pseudo labels produce stable self-learning signals in numerous multi-modal test-time adaptation scenarios for 3D semantic segmentation. Visit our project website at https://www.nec-labs.com/~mas/MM-TTA

We propose a novel method on refining cross-person gaze prediction task with eye/face images only by explicitly modelling the person-specific differences. Specifically, we first assume that we can obtain some initial gaze prediction results with existing method, which we refer to as InitNet, and then introduce three modules, the Validity Module (VM), Self-Calibration (SC) and Person-specific Transform (PT) module. By predicting the reliability of current eye/face images, VM is able to identify invalid samples, e.g. eye blinking images, and reduce their effects in modelling process. SC and PT module then learn to compensate for the differences on valid samples only. The former models the translation offsets by bridging the gap between initial predictions and dataset-wise distribution. And the later learns more general person-specific transformation by incorporating the information from existing initial predictions of the same person. We validate our ideas on three publicly available datasets, EVE, XGaze, and MPIIGaze dataset. We demonstrate that our proposed method outperforms the SOTA methods significantly on all of them, e.g. respectively 21.7%, 36.0%, and 32.9% relative performance improvements. We are the winner of the GAZE 2021 EVE Challenge and our code can be found here https://github.com/bjj9/EVE_SCPT.

Trajectory prediction is a safety-critical tool for autonomous vehicles to plan and execute actions. Our work addresses two key challenges in trajectory prediction, learning multimodal outputs, and better predictions by imposing constraints using driving knowledge. Recent methods have achieved strong performances using Multi-Choice Learning objectives like winner-takes-all (WTA) or best-of-many. But the impact of those methods in learning diverse hypotheses is under-studied as such objectives highly depend on their initialization for diversity. As our first contribution, we propose a novel Divide-And-Conquer (DAC) approach that acts as a better initialization technique to WTA objective, resulting in diverse outputs without any spurious modes. Our second contribution is a novel trajectory prediction framework called ALAN that uses existing lane centerlines as anchors to provide trajectories constrained to the input lanes. Our framework provides multi-agent trajectory outputs in a forward pass by capturing interactions through hypercolumn descriptors and incorporating scene information in the form of rasterized images and per-agent lane anchors. Experiments on synthetic and real data show that the proposed DAC captures the data distribution better compare to other WTA family of objectives. Further, we show that our ALAN approach provides on par or better performance with SOTA methods evaluated on Nuscenes urban driving benchmark.

We address the challenging task of occlusion-aware indoor 3D scene understanding. We represent scenes by a set of planes, where each one is defined by its normal, offset and two masks outlining (i) the extent of the visible part and (ii) the full region that consists of both visible and occluded parts of the plane. We infer these planes from a single input image with a novel neural network architecture. It consists of a two-branch category-specific module that aims to predict layout and objects of the scene separately so that different types of planes can be handled better. We also introduce a novel loss function based on plane warping that can leverage multiple views at training time for improved occlusion-aware reasoning. In order to train and evaluate our occlusion-reasoning model, we use the ScanNet dataset and propose (i) a strategy to automatically extract ground truth for both visible and hidden regions and (ii) a new evaluation metric that specifically focuses on the prediction in hidden regions. We empirically demonstrate that our proposed approach can achieve higher accuracy for occlusion reasoning compared to competitive baselines on the ScanNet dataset, e.g. 42.65% relative improvement on hidden regions.

In this paper, we address the problem of inferring the layout of complex road scenes from video sequences. To this end, we formulate it as a top-view road attributes prediction problem and our goal is to predict these attributes for each frame both accurately and consistently. In contrast to prior work, we exploit the following three novel aspects: leveraging camera motions in videos, including context cues and incorporating long-term video information. Specifically, we introduce a model that aims to enforce prediction consistency in videos. Our model consists of one LSTM and one Feature Transform Module (FTM). The former implicitly incorporates the consistency constraint with its hidden states, and the latter explicitly takes the camera motion into consideration when aggregating information along videos. Moreover, we propose to incorporate context information by introducing road participants, e.g. objects, into our model. When the entire video sequence is available, our model is also able to encode both local and global cues, e.g. information from both past and future frames. Experiments on two data sets show that: (1) Incorporating either global or contextual cues improves the prediction accuracy and leveraging both gives the best performance. (2) Introducing the LSTM and FTM modules improves the prediction consistency in videos. (3) The proposed method outperforms the SOTA by a large margin.

We propose an active learning approach for transferring representations across domains. Our approach, active adversarial domain adaptation (AADA), explores a duality between two related problems: adversarial domain alignment and importance sampling for adapting models across domains. The former uses a domain discriminative model to align domains, while the latter utilizes the model to weigh samples to account for distribution shifts. Specifically, our importance weight promotes unlabeled samples with large uncertainty in classification and diversity compared to la-beled examples, thus serving as a sample selection scheme for active learning. We show that these two views can be unified in one framework for domain adaptation and transfer learning when the source domain has many labeled examples while the target domain does not. AADA provides significant improvements over fine-tuning based approaches and other sampling methods when the two domains are closely related. Results on challenging domain adaptation tasks such as object detection demonstrate that the advantage over baseline approaches is retained even after hundreds of examples being actively annotated.

In this paper, we address the problem of inferring the layout of complex road scenes given a single camera as input. To achieve that, we first propose a novel parameterized model of road layouts in a top-view representation, which is not only intuitive for human visualization but also provides an interpretable interface for higher-level decision making. Moreover, the design of our top-view scene model allows for efficient sampling and thus generation of large-scale simulated data, which we leverage to train a deep neural network to infer our scene model's parameters. Specifically, our proposed training procedure uses supervised domain-adaptation techniques to incorporate both simulated as well as manually annotated data. Finally, we design a Conditional Random Field (CRF) that enforces coherent predictions for a single frame and encourages temporal smoothness among video frames. Experiments on two public data sets show that: (1) Our parametric top-view model is representative enough to describe complex road scenes; (2) The proposed method outperforms baselines trained on manually-annotated or simulated data only, thus getting the best of both; (3) Our CRF is able to generate temporally smoothed while semantically meaningful results.

Annotating human poses in realistic scenes is very time consuming, yet necessary for training human pose estimators. We propose to address this problem in an active learning framework, which alternates between requesting the most useful annotations among a large set of unlabelled images, and re-training the pose estimator. To this end, (1) we propose an uncertainty estimator specific for body joint predictions, which takes into account the spatial distribution of the responses of the current pose estimator on the unlabelled images; (2) we propose a dynamic combination of influence and uncertainty cues, where their weights vary during the active learning process according to the reliability of the current pose estimator; (3) we introduce a computer assisted annotation interface, which reduces the time necessary for a human annotator to click on a joint by discretizing the image into regions generated by the current pose estimator. Experiments using the MPII and LSP datasets with both simulated and real annotators show that (1) the proposed active selection scheme outperforms several baselines; (2) our computer-assisted interface can further reduce annotation effort; and (3) our technique can further improve the performance of a pose estimator even when starting from an already strong one.

We address the problem of integrating object reasoning with supervoxel labeling in multiclass semantic video segmentation. To this end, we first propose an object-augmented dense CRF in spatio-temporal domain, which captures long-range dependency between supervoxels, and imposes consistency between object and supervoxel labels. We develop an efficient mean field inference algorithm to jointly infer the supervoxel labels, object activations and their occlusion relations for a moderate number of object hypotheses. To scale up our method, we adopt an active inference strategy to improve the efficiency, which adaptively selects object subgraphs in the object-augmented dense CRF. We formulate the problem as a Markov Decision Process, which learns an approximate optimal policy based on a reward of accuracy improvement and a set of well-designed model and input features. We evaluate our method on three publicly available multiclass video semantic segmentation datasets and demonstrate superior efficiency and accuracy.

We tackle the problem of semantic segmentation of dynamic scene in video sequences. We propose to incorporate foreground object information into pixel labeling by jointly reasoning semantic labels of super-voxels, object instance tracks and geometric relations between objects. We take an exemplar approach to object modeling by using a small set of object annotations and exploring the temporal consistency of object motion. After generating a set of moving object hypotheses, we design a CRF framework that jointly models the super voxel and object instances. The optimal semantic labeling is inferred by the MAP estimation of the model, which is solved by a single move-making based optimization procedure. We demonstrate the effectiveness of our method on three public datasets and show that our model can achieve superior or comparable results than the state of-the-art with less object-level supervision.

We address the problem of geometric and semantic consistent video segmentation for outdoor scenes. With no assumption on camera movement, we jointly model the semantic-geometric class of spatio-temporal regions (supervoxels) and geometric scene layout in each frame. Our main contribution is to propose a stage scene model to efficiently capture the dependency between the semantic and geometric labels. We build a unified CRF model on supervoxel labels and stage parameters, and design an alternating inference algorithm to minimize the resulting energy function. We also extend smoothing based on hierarchical image segmentation to spatio-temporal setting and show it achieves better performance than a pairwise random field model. Our method is evaluated on the CamVid dataset and achieves state-of-the-art per-pixel as well as per-class accuracy in predicting both semantic and geometric labels.