Vision-and-Language Navigation (VLN) agents are tasked with navigating an unseen environment using natural language instructions. In this work, we study if visual representations of sub-goals implied by the instructions can serve as navigational cues and lead to increased navigation performance. To synthesize these visual representations or "imaginations", we leverage a text-to-image diffusion model on landmark references contained in segmented instructions. These imaginations are provided to VLN agents as an added modality to act as landmark cues and an auxiliary loss is added to explicitly encourage relating these with their corresponding referring expressions. Our findings reveal an increase in success rate (SR) of ~1 point and up to ~0.5 points in success scaled by inverse path length (SPL) across agents. These results suggest that the proposed approach reinforces visual understanding compared to relying on language instructions alone. Code and data for our work can be found at https://www.akhilperincherry.com/VLN-Imagine-website/.

Multimodal large language models (MLLMs) are increasingly used to evaluate text-to-image (TTI) generation systems, providing automated judgments based on visual and textual context. However, these "judge" models often suffer from biases, overconfidence, and inconsistent performance across diverse image domains. While prompt ensembling has shown promise for mitigating these issues in unimodal, text-only settings, our experiments reveal that standard ensembling methods fail to generalize effectively for TTI tasks. To address these limitations, we propose a new multimodal-aware method called Multimodal Mixture-of-Bayesian Prompt Ensembles (MMB). Our method uses a Bayesian prompt ensemble approach augmented by image clustering, allowing the judge to dynamically assign prompt weights based on the visual characteristics of each sample. We show that MMB improves accuracy in pairwise preference judgments and greatly enhances calibration, making it easier to gauge the judge's true uncertainty. In evaluations on two TTI benchmarks, HPSv2 and MJBench, MMB outperforms existing baselines in alignment with human annotations and calibration across varied image content. Our findings highlight the importance of multimodal-specific strategies for judge calibration and suggest a promising path forward for reliable large-scale TTI evaluation.

Dormant tree pruning is laborintensive but essential to maintaining modern highly-productive fruit orchards. In this work we present a closed-loop visuomotor controller for robotic pruning. The controller guides the cutter through a cluttered tree environment to reach a specified cut point and ensures the cutters are perpendicular to the branch. We train the controller using a novel orchard simulation that captures the geometric distribution of branches in a target apple orchard configuration. Unlike traditional methods requiring full 3D reconstruction, our controller uses just optical flow images from a wrist-mounted camera. We deploy our learned policy in simulation and the real-world for an example V-Trellis envy tree with zero-shot transfer, achieving a ~30% success rate - approximately half the performance of an oracle planner.

We propose a novel approach to learn relational policies for classical planning based on learning to rank actions. We introduce a new graph representation that explicitly captures action information and propose a Graph Neural Network (GNN) architecture augmented with Gated Recurrent Units (GRUs) to learn action rankings. Unlike value-function based approaches that must learn a globally consistent function, our action ranking method only needs to learn locally consistent ranking. Our model is trained on data generated from small problem instances that are easily solved by planners and is applied to significantly larger instances where planning is computationally prohibitive. Experimental results across standard planning benchmarks demonstrate that our action-ranking approach not only achieves better generalization to larger problems than those used in training but also outperforms multiple baselines (value function and action ranking) methods in terms of success rate and plan quality.

Assistive embodied agents that can be instructed in natural language to perform tasks in open-world environments have the potential to significantly impact labor tasks like manufacturing or in-home care - benefiting the lives of those who come to depend on them. In this work, we consider how this benefit might be hijacked by local modifications in the appearance of the agent's operating environment. Specifically, we take the popular Vision-and-Language Navigation (VLN) task as a representative setting and develop a whitebox adversarial attack that optimizes a 3D attack object's appearance to induce desired behaviors in pretrained VLN agents that observe it in the environment. We demonstrate that the proposed attack can cause VLN agents to ignore their instructions and execute alternative actions after encountering the attack object - even for instructions and agent paths not considered when optimizing the attack. For these novel settings, we find our attacks can induce early-termination behaviors or divert an agent along an attacker-defined multi-step trajectory. Under both conditions, environmental attacks significantly reduce agent capabilities to successfully follow user instructions.

Referring expressions for visual objects often include descriptions of relative spatial arrangements to other objects - e.g. “to the right of” - that depend on the point of view of the speaker. In 2D referring expression tasks, this view-point is captured unambiguously in the image. However, grounding expressions with such spatial language in 3D without viewpoint annotations can be ambiguous. In this paper, we investigate the significance of viewpoint information in 3D visual grounding - introducing a model that explicitly predicts the speaker's viewpoint based on the referring expression and scene. We pretrain this model on a synthetically generated dataset that provides viewpoint annotations and then finetune on 3D referring expression datasets. Further, we introduce an auxiliary uniform object representation loss to encourage viewpoint invariance in learned object representations. We find that our proposed ViewPoint Prediction Network (VPP-Net) achieves state-of-the-art performance on ScanRefer, SR3D, and NR3D - improving [email protected] by 1.06%, 0.60%, and 2.00% respectively compared to prior work.

Recent dataset deduplication techniques have demonstrated that content-aware dataset pruning can dramatically reduce the cost of training Vision-Language Pre-trained (VLP) models without significant performance losses compared to training on the original dataset. These results have been based on pruning commonly used image-caption datasets collected from the web - datasets that are known to harbor harmful social biases that may then be codified in trained models. In this work, we evaluate how deduplication affects the prevalence of these biases in the resulting trained models and introduce an easy-to-implement modification to the recent SemDeDup algorithm that can reduce the negative effects that we observe. When examining CLIP-style models trained on deduplicated variants of LAION-400M, we find our proposed FairDeDup algorithm consistently leads to improved fairness metrics over SemDeDup on the FairFace and FACET datasets while maintaining zero-shot performance on CLIP benchmarks.

This paper unifies commonly used accelerated stochastic gradient methods (Polyak’s Heavy Ball, Nesterov’s Accelerated Gradient and Adaptive Moment Estimation (Adam)) as specific cases of a general lowpass regularized learning framework, the Automatic Stochastic Gradient Method (AutoSGM). For AutoSGM, we derive an optimal iteration-dependent learning rate function and realize an approximation. Adam is also an approximation of this optimal approach that replaces the iteration-dependent learning-rate with a constant. Empirical results on deep neural networks comparing the learning behavior of AutoSGM equipped with this iteration-dependent learning-rate algorithm demonstrate fast learning behavior, robustness to the initial choice of the learning rate, and can tune an initial constant learning-rate in applications where a good constant learning rate approximation is unknown.

Visual control policies can encounter significant performance degradation when visual conditions like lighting or camera position differ from those seen during training – often exhibiting sharp declines in capability even for minor differences. In this work, we examine robustness to a suite of these types of visual changes for RGB-D and point cloud based visual control policies. To perform these experiments on both model-free and model-based reinforcement learners, we introduce a novel Point Cloud World Model (PCWM) and point cloud based control policies. Our experiments show that policies that explicitly encode point clouds are significantly more robust than their RGB-D counterparts. Further, we find our proposed PCWM significantly outperforms prior works in terms of sample efficiency during training. Taken together, these results suggest reasoning about the 3D scene through point clouds can improve performance, reduce learning time, and increase robustness for robotic learners. Code: https://github.com/pvskand/pcwm

We study the problem of natural language-based control of dynamic bipedal locomotion from the perspective of operational robustness and hardware safety. Existing work on natural language-based robot control has focused on episodic command execution for stable robot platforms, such as fixed-based manipulators in table-top scenarios. These scenarios feature non-overlapping phases of instruction and execution, with execution mishaps usually posing no threat to the robot safety. This allows for non-trivial failure rates to be acceptable. In contrast, our work involves indistinguishable instruction and execution stages for a dynamically unstable robot where execution failures can harm the robot. For example, interrupting a bipedal robot with a new instruction in certain states may cause it to fall. Our first contribution is to design and train a natural language-based controller for the bipedal robot Cassie that can take in new language commands at any time. Our second contribution is to introduce a protocol for evaluating the robustness to interruptions of such controllers and evaluating the learned controller in simulation under different interruption distributions. Our third contribution is to learn a detector for interruptions that are likely to lead to failure and to integrate that detector into a failure mitigation strategy. Overall, our results show that interruptions can lead to non-trivial failure rates for the original controller and that the proposed mitigation strategy can help to significantly reduce that rate.

When faced with an out-of-distribution (OOD) question or image, visual question answering (VQA) systems may provide unreliable answers. If relied on by real users or secondary systems, these failures may range from annoying to potentially endangering. Detecting OOD samples in single-modality settings is well-studied; however, limited attention has been paid to vision-and-language settings. In this work, we examine the question of OOD detection in the multimodal VQA task and benchmark a suite of approaches to identify OOD image-question pairs. In our experiments, we leverage popular VQA datasets to benchmark detection performance across a range of difficulties. We also produce composite datasets to examine impacts of individual modalities and of image-question agreement. Our results show that answer confidence alone is often a poor signal and that methods based on image-based question generation or examining model attention can lead to significantly better results. We find detecting ungrounded image-question pairs and small shifts in image distribution remain challenging.

We present Iterative Vision-and-Language Navigation (IVLN), a paradigm for evaluating language-guided agents navigating in a persistent environment over time. Existing Vision-and-Language Navigation (VLN) benchmarks erase the agent's memory at the beginning of every episode, testing the ability to perform cold-start navigation with no prior information. However, deployed robots occupy the same environment for long periods of time. The IVLN paradigm addresses this disparity by training and evaluating VLN agents that maintain memory across tours of scenes that consist of up to 100 ordered instruction-following Room-to-Room (R2R) episodes, each defined by an individual language instruction and a target path. We present discrete and continuous Iterative Room-to-Room (IR2R) benchmarks comprising about 400 tours each in 80 indoor scenes. We find that extending the implicit memory of high-performing transformer VLN agents is not sufficient for IVLN, but agents that build maps can benefit from environment persistence, motivating a renewed focus on map-building agents in VLN.

To be successful, Vision-and-Language Navigation (VLN) agents must be able to ground instructions to actions based on their surroundings. In this work, we develop a methodology to study agent behavior on a skill-specific basis - examining how well existing agents ground instructions about stopping, turning, and moving towards specified objects or rooms. Our approach is based on generating skill-specific interventions and measuring changes in agent predictions. We present a detailed case study analyzing the behavior of a recent agent and then compare multiple agents in terms of skill-specific competency scores. This analysis suggests that biases from training have lasting effects on agent behavior and that existing models are able to ground simple referring expressions. Our comparisons between models show that skill-specific scores correlate with improvements in overall VLN task performance.

Images are a convenient way to specify which particular object instance an embodied agent should navigate to. Solving this task requires semantic visual reasoning and exploration of unknown environments. We present a system that can perform this task in both simulation and the real world. Our modular method solves sub-tasks of exploration, goal instance re-identification, goal localization, and local navigation. We re-identify the goal instance in egocentric vision using feature-matching and localize the goal instance by projecting matched features to a map. Each sub-task is solved using off-the-shelf components requiring zero fine-tuning. On the HM3D InstanceImageNav benchmark, this system outperforms a baseline end-to-end RL policy 7x and a state-of-the-art ImageNav model 2.3x (56% vs . 25% success). We deploy this system to a mobile robot platform and demonstrate effective real-world performance, achieving an 88% success rate across a home and an office environment.

Recent work in vision-and-language demonstrates that large-scale pretraining can learn generalizable models that are efficiently transferable to downstream tasks. While this may improve dataset-scale aggregate metrics, analyzing performance around hand-crafted subgroups targeting specific bias dimensions reveals systemic undesirable behaviors. However, this subgroup analysis is frequently stalled by annotation efforts, which require extensive time and resources to collect the necessary data. Prior art attempts to automatically discover subgroups to circumvent these constraints but typically leverages model behavior on existing task-specific annotations and rapidly degrades on more complex inputs beyond "tabular" data, none of which study vision-and-language models. This paper presents VLSlice, an interactive system enabling user-guided discovery of coherent representation-level subgroups with consistent visiolinguistic behavior, denoted as vision-and-language slices, from unlabeled image sets. We show that VLSlice enables users to quickly generate diverse high-coherency slices in a user study (n=22) and release the tool publicly1.

Users often want to augment and enrich entities in their datasets with relevant information from external data sources. As many external sources are accessible only via keyword-search interfaces, a user usually has to manually formulate a keyword query that extract relevant information for each entity. This approach is challenging as many data sources contain numerous tuples, only a small fraction of which may contain entity-relevant information. Furthermore, different datasets may represent the same information in distinct forms and under different terms (e.g., different data source may use different names to refer to the same person). In such cases, it is difficult to formulate a query that precisely retrieves information relevant to an entity. Current methods for information enrichment mainly rely on lengthy and resource-intensive manual effort to formulate queries to discover relevant information. However, in increasingly many settings, it is important for users to get initial answers quickly and without substantial investment in resources (such as human attention). We propose a progressive approach to discovering entity-relevant information from external sources with minimal expert intervention. It leverages end users' feedback to progressively learn how to retrieve information relevant to each entity in a dataset from external data sources. Our empirical evaluation shows that our approach learns accurate strategies to deliver relevant information quickly.

We study the task of semantic mapping – specifically, an embodied agent (a robot or an egocentric AI assistant) is given a tour of a new environment and asked to build an allocentric top-down semantic map (‘what is where?’) from egocentric observations of an RGB-D camera with known pose (via localization sensors). Importantly, our goal is to build neural episodic memories and spatio-semantic representations of 3D spaces that enable the agent to easily learn subsequent tasks in the same space – navigating to objects seen during the tour (‘Find chair’) or answering questions about the space (‘How many chairs did you see in the house?’). Towards this goal, we present Semantic MapNet (SMNet), which consists of: (1) an Egocentric Visual Encoder that encodes each egocentric RGB-D frame, (2) a Feature Projector that projects egocentric features to appropriate locations on a floor-plan, (3) a Spatial Memory Tensor of size floor-plan length×width×feature-dims that learns to accumulate projected egocentric features, and (4) a Map Decoder that uses the memory tensor to produce semantic top-down maps. SMNet combines the strengths of (known) projective camera geometry and neural representation learning. On the task of semantic mapping in the Matterport3D dataset, SMNet significantly outperforms competitive baselines by 4.01−16.81% (absolute) on mean-IoU and 3.81−19.69% (absolute) on Boundary-F1 metrics. Moreover, we show how to use the spatio-semantic allocentric representations build by SMNet for the task of ObjectNav and Embodied Question Answering. Project page: https://vincentcartillier.github.io/smnet.html.

Multilingual Neural Machine Translation (NMT) enables one model to serve all translation directions, including ones that are unseen during training, i.e. zero-shot translation.Despite being theoretically attractive, current models often produce low quality translations -commonly failing to even produce outputs in the right target language.In this work, we observe that off-target translation is dominant even in strong multilingual systems, trained on massive multilingual corpora.To address this issue, we propose a joint approach to regularize NMT models at both representation-level and gradient-level.At the representation level, we leverage an auxiliary target language prediction task to regularize decoder outputs to retain information about the target language.At the gradient level, we leverage a small amount of direct data (in thousands of sentence pairs) to regularize model gradients.Our results demonstrate that our approach is highly effective in both reducing off-target translation occurrences and improving zero-shot translation performance by +5.59 and +10.38 BLEU on WMT and OPUS datasets respectively.Moreover, experiments show that our method also works well when the small amount of direct data is not available.1

Little inquiry has explicitly addressed the role of action spaces in language-guided visual navigation – either in terms of its effect on navigation success or the efficiency with which a robotic agent could execute the resulting trajectory. Building on the recently released VLN-CE [24] setting for instruction following in continuous environments, we develop a class of language-conditioned waypoint prediction networks to examine this question. We vary the expressivity of these models to explore a spectrum between low-level actions and continuous waypoint prediction. We measure task performance and estimated execution time on a profiled LoCoBot [1] robot. We find more expressive models result in simpler, faster to execute trajectories, but lower-level actions can achieve better navigation metrics by approximating shortest paths better. Further, our models outperform prior work in VLN-CE and set a new state-of-the-art on the public leaderboard – increasing success rate by 4% with our best model on this challenging task.

Can general-purpose neural models learn to navigate? For PointGoal navigation (‘go to Δx, Δy’), the answer is a clear ‘yes’ – mapless neural models composed of task-agnostic components (CNNs and RNNs) trained with large-scale model-free reinforcement learning achieve near-perfect performance [27]. However, for ObjectGoal navigation (‘find a TV’), this is an open question; one we tackle in this paper. The current best-known result on ObjectNav with general-purpose models is 6% success rate [25].First, we show that the key problem is overfitting. Large-scale training results in 94% success rate on training environments and only 8% in validation. We observe that this stems from agents memorizing environment layouts during training – sidestepping the need for exploration and directly learning shortest paths to nearby goal objects. We show that this is a natural consequence of optimizing for the task metric (which in fact penalizes exploration), is enabled by powerful observation encoders, and is possible due to the finite set of training environment configurations.Informed by our findings, we introduce Treasure Hunt Data Augmentation (THDA) to address overfitting in ObjectNav. THDA inserts 3D scans of household objects at arbitrary scene locations and uses them as ObjectNav goals – augmenting and greatly expanding the set of training layouts. Taken together with our other proposed changes, we improve the state of art on the Habitat ObjectGoal Navigation benchmark by 90% (from 14% success rate to 27%) and path efficiency by 48% (from 7.5 SPL to 11.1 SPL).

Natural language instructions for visual navigation often use scene descriptions (e.g., bedroom) and object references (e.g., green chairs) to provide a breadcrumb trail to a goal location. This work presents a transformer-based vision-and-language navigation (VLN) agent that uses two different visual encoders -- a scene classification network and an object detector -- which produce features that match these two distinct types of visual cues. In our method, scene features contribute high-level contextual information that supports object-level processing. With this design, our model is able to use vision-and-language pretraining (i.e., learning the alignment between images and text from large-scale web data) to substantially improve performance on the Room-to-Room (R2R) and Room-Across-Room (RxR) benchmarks. Specifically, our approach leads to improvements of 1.8% absolute in SPL on R2R and 3.7% absolute in SR on RxR. Our analysis reveals even larger gains for navigation instructions that contain six or more object references, which further suggests that our approach is better able to use object features and align them to references in the instructions.

Top-performing approaches to embodied AI tasks like point-goal navigation often rely on training agents via reinforcement learning over tens of millions (or even billions) of experiential steps - learning neural agents that map directly from visual observations to actions. In this work, we question whether these extreme training durations are necessary or if they are simply due to the difficulty of learning visual representations purely from task reward. We examine the task of point-goal navigation in photorealistic environments and introduce three auxiliary tasks that encourage learned representations to capture key elements of the task - local scene geometry, transition dynamics of the environment, and progress towards the goal. Importantly, these can be evaluated independent of task performance and provide strong supervision for representation learning. Our auxiliary tasks are simple to implement and rely on supervision already present in simulators commonly used for point-goal navigation. Applying our auxiliary losses to agents from prior works, we observe a >4 × improvement in sample efficiency - in 17 million steps, our augmented agents outperforms state-of-the-art agents trained for 72 million steps.

Much of vision-and-language research focuses on a small but diverse set of independent tasks and supporting datasets often studied in isolation; however, the visually-grounded language understanding skills required for success at these tasks overlap significantly. In this work, we investigate these relationships between vision-and-language tasks by developing a large-scale, multi-task model. Our approach culminates in a single model on 12 datasets from four broad categories of task including visual question answering, caption-based image retrieval, grounding referring expressions, and multimodal verification. Compared to independently trained single-task models, this represents a reduction from approximately 3 billion parameters to 270 million while simultaneously improving performance by 2.05 points on average across tasks. We use our multi-task framework to perform in-depth analysis of the effect of joint training diverse tasks. Further, we show that finetuning task-specific models from our single multi-task model can lead to further improvements, achieving performance at or above the state-of-the-art.

We present WHERE ARE YOU? (WAY), a dataset of ~6k dialogs in which two humans – an Observer and a Locator – complete a cooperative localization task. The Observer is spawned at random in a 3D environment and can navigate from first-person views while answering questions from the Locator. The Locator must localize the Observer in a detailed top-down map by asking questions and giving instructions. Based on this dataset, we define three challenging tasks: Localization from Embodied Dialog or LED (localizing the Observer from dialog history), Embodied Visual Dialog (modeling the Observer), and Cooperative Localization (modeling both agents). In this paper, we focus on the LED task – providing a strong baseline model with detailed ablations characterizing both dataset biases and the importance of various modeling choices. Our best model achieves 32.7% success at identifying the Observer's location within 3m in unseen buildings, vs. 70.4% for human Locators.

Can we develop visually grounded dialog agents that can efficiently adapt to new tasks without forgetting how to talk to people? Such agents could leverage a larger variety of existing data to generalize to a new task, minimizing expensive data collection and annotation. In this work, we study a setting we call "Dialog without Dialog", which requires agents to develop visually grounded dialog models that can adapt to new tasks without language level supervision. By factorizing intention and language, our model minimizes linguistic drift after fine-tuning for new tasks. We present qualitative results, automated metrics, and human studies that all show our model can adapt to new tasks and maintain language quality. Baselines either fail to perform well at new tasks or experience language drift, becoming unintelligible to humans. Code has been made available at: https://github.com/mcogswell/dialogwithoutdialog.

Imitation learning is a popular approach for teaching motor skills to robots. However, most approaches focus on extracting policy parameters from execution traces alone (i.e., motion trajectories and perceptual data). No adequate communication channel exists between the human expert and the robot to describe critical aspects of the task, such as the properties of the target object or the intended shape of the motion. Motivated by insights into the human teaching process, we introduce a method for incorporating unstructured natural language into imitation learning. At training time, the expert can provide demonstrations along with verbal descriptions in order to describe the underlying intent (e.g., "go to the large green bowl"). The training process then interrelates these two modalities to encode the correlations between language, perception, and motion. The resulting language-conditioned visuomotor policies can be conditioned at runtime on new human commands and instructions, which allows for more fine-grained control over the trained policies while also reducing situational ambiguity. We demonstrate in a set of simulation experiments how our approach can learn language-conditioned manipulation policies for a seven-degree-of-freedom robot arm and compare the results to a variety of alternative methods.

We introduce the task of scene-aware dialog. Our goal is to generate a complete and natural response to a question about a scene, given video and audio of the scene and the history of previous turns in the dialog. To answer successfully, agents must ground concepts from the question in the video while leveraging contextual cues from the dialog history. To benchmark this task, we introduce the Audio Visual Scene-Aware Dialog (AVSD) Dataset. For each of more than 11,000 videos of human actions from the Charades dataset, our dataset contains a dialog about the video, plus a final summary of the video by one of the dialog participants. We train several baseline systems for this task and evaluate the performance of the trained models using both qualitative and quantitative metrics. Our results indicate that models must utilize all the available inputs (video, audio, question, and dialog history) to perform best on this dataset.

To help bridge the gap between internet vision-style problems and the goal of vision for embodied perception we instantiate a large-scale navigation task -- Embodied Question Answering [1] in photo-realistic environments (Matterport 3D). We thoroughly study navigation policies that utilize 3D point clouds, RGB images, or their combination. Our analysis of these models reveals several key findings. We find that two seemingly naive navigation baselines, forward-only and random, are strong navigators and challenging to outperform, due to the specific choice of the evaluation setting presented by [1]. We find a novel loss-weighting scheme we call Inflection Weighting to be important when training recurrent models for navigation with behavior cloning and are able to out perform the baselines with this technique. We find that point clouds provide a richer signal than RGB images for learning obstacle avoidance, motivating the use (and continued study) of 3D deep learning models for embodied navigation.

Arijit Ray, Karan Sikka, Ajay Divakaran, Stefan Lee, Giedrius Burachas. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Image captioning models have achieved impressive results on datasets containing limited visual concepts and large amounts of paired image-caption training data. However, if these models are to ever function in the wild, a much larger variety of visual concepts must be learned, ideally from less supervision. To encourage the development of image captioning models that can learn visual concepts from alternative data sources, such as object detection datasets, we present the first large-scale benchmark for this task. Dubbed `nocaps', for novel object captioning at scale, our benchmark consists of 166,100 human-generated captions describing 15,100 images from the Open Images validation and test sets. The associated training data consists of COCO image-caption pairs, plus Open Images image-level labels and object bounding boxes. Since Open Images contains many more classes than COCO, nearly 400 object classes seen in test images have no or very few associated training captions (hence, nocaps). We extend existing novel object captioning models to establish strong baselines for this benchmark and provide analysis to guide future work.

Many vision and language models suffer from poor visual grounding -- often falling back on easy-to-learn language priors rather than basing their decisions on visual concepts in the image. In this work, we propose a generic approach called Human Importance-aware Network Tuning (HINT) that effectively leverages human demonstrations to improve visual grounding. HINT encourages deep networks to be sensitive to the same input regions as humans. Our approach optimizes the alignment between human attention maps and gradient-based network importances -- ensuring that models learn not just to look at but rather rely on visual concepts that humans found relevant for a task when making predictions. We apply HINT to Visual Question Answering and Image Captioning tasks, outperforming top approaches on splits that penalize over-reliance on language priors (VQA-CP and robust captioning) using human attention demonstrations for just 6% of the training data.

In this work, we develop a technique to produce counterfactual visual explanations. Given a ‘query’ image $I$ for which a vision system predicts class $c$, a counterfactual visual explanation identifies how $I$ could change such that the system would output a different specified class $c’$. To do this, we select a ‘distractor’ image $I’$ that the system predicts as class $c’$ and identify spatial regions in $I$ and $I’$ such that replacing the identified region in $I$ with the identified region in $I’$ would push the system towards classifying $I$ as $c’$. We apply our approach to multiple image classification datasets generating qualitative results showcasing the interpretability and discriminativeness of our counterfactual explanations. To explore the effectiveness of our explanations in teaching humans, we present machine teaching experiments for the task of fine-grained bird classification. We find that users trained to distinguish bird species fare better when given access to counterfactual explanations in addition to training examples.

Many sequence prediction tasks admit multiple correct outputs and so, it is often useful to decode a set of outputs that maximize some task-specific set-level metric. However, retooling standard sequence prediction procedures tailored towards predicting the single best output leads to the decoding of sets containing very similar sequences; failing to capture the variation in the output space. To address this, we propose $\nabla$BS, a trainable decoding procedure that outputs a set of sequences, highly valued according to the metric. Our method tightly integrates the training and decoding phases and further allows for the optimization of the task-specific metric addressing the shortcomings of standard sequence prediction. Further, we discuss the trade-offs of commonly used set-level metrics and motivate a new set-level metric that naturally evaluates the notion of “capturing the variation in the output space”. Finally, we show results on the image captioning task and find that our model outperforms standard techniques and natural ablations.

We propose a new class of probabilistic neural-symbolic models, that have symbolic functional programs as a latent, stochastic variable. Instantiated in the context of visual question answering, our probabilistic formulation offers two key conceptual advantages over prior neural-symbolic models for VQA. Firstly, the programs generated by our model are more understandable while requiring less number of teaching examples. Secondly, we show that one can pose counterfactual scenarios to the model, to probe its beliefs on the programs that could lead to a specified answer given an image. Our results on the CLEVR and SHAPES datasets verify our hypotheses, showing that the model gets better program (and answer) prediction accuracy even in the low data regime, and allows one to probe the coherence and consistency of reasoning performed.

A visually-grounded navigation instruction can be interpreted as a sequence of expected observations and actions an agent following the correct trajectory would encounter and perform. Based on this intuition, we formulate the problem of finding the goal location in Vision-and-Language Navigation (VLN) within the framework of Bayesian state tracking - learning observation and motion models conditioned on these expectable events. Together with a mapper that constructs a semantic spatial map on-the-fly during navigation, we formulate an end-to-end differentiable Bayes filter and train it to identify the goal by predicting the most likely trajectory through the map according to the instructions. The resulting navigation policy constitutes a new approach to instruction following that explicitly models a probability distribution over states, encoding strong geometric and algorithmic priors while enabling greater explainability. Our experiments show that our approach outperforms a strong LingUNet baseline when predicting the goal location on the map. On the full VLN task, i.e. navigating to the goal location, our approach achieves promising results with less reliance on navigation constraints.

We present ViLBERT (short for Vision-and-Language BERT), a model for learning task-agnostic joint representations of image content and natural language. We extend the popular BERT architecture to a multi-modal two-stream model, processing both visual and textual inputs in separate streams that interact through co-attentional transformer layers. We pretrain our model through two proxy tasks on the large, automatically collected Conceptual Captions dataset and then transfer it to multiple established vision-and-language tasks -- visual question answering, visual commonsense reasoning, referring expressions, and caption-based image retrieval -- by making only minor additions to the base architecture. We observe significant improvements across tasks compared to existing task-specific models -- achieving state-of-the-art on all four tasks. Our work represents a shift away from learning groundings between vision and language only as part of task training and towards treating visual grounding as a pretrainable and transferable capability.

We introduce the task of Visual Dialog, which requires an AI agent to hold a meaningful dialog with humans in natural, conversational language about visual content. Specifically, given an image, a dialog history, and a question about the image, the agent has to ground the question in image, infer context from history, and answer the question accurately. Visual Dialog is disentangled enough from a specific downstream task so as to serve as a general test of machine intelligence, while being sufficiently grounded in vision to allow objective evaluation of individual responses and benchmark progress. We develop a novel two-person real-time chat data-collection protocol to curate a large-scale Visual Dialog dataset (VisDial). VisDial v0.9 has been released and consists of$\sim$1.2M dialog question-answer pairs from 10-round, human-human dialogs grounded in$\sim$120k images from the COCO dataset. We introduce a family of neural encoder-decoder models for Visual Dialog with 3 encoders—Late Fusion, Hierarchical Recurrent Encoder and Memory Network (optionally with attention over image features)—and 2 decoders (generative and discriminative), which outperform a number of sophisticated baselines. We propose a retrieval-based evaluation protocol for Visual Dialog where the AI agent is asked to sort a set of candidate answers and evaluated on metrics such as mean-reciprocal-rank and recall$@k$of human response. We quantify the gap between machine and human performance on the Visual Dialog task via human studies. Putting it all together, we demonstrate the first ‘visual chatbot’! Our dataset, code, pretrained models and visual chatbot are available onhttps://visualdialog.org.

A single image captures the appearance and position of multiple entities in a scene as well as their complex interactions. As a consequence, natural language grounded in visual contexts tends to be diverse---with utterances differing as focus shifts to specific objects, interactions, or levels of detail. Recently, neural sequence models such as RNNs and LSTMs have been employed to produce visually-grounded language. Beam Search, the standard work-horse for decoding sequences from these models, is an approximate inference algorithm that decodes the top-B sequences in a greedy left-to-right fashion. In practice, the resulting sequences are often minor rewordings of a common utterance, failing to capture the multimodal nature of source images. To address this shortcoming, we propose Diverse Beam Search (DBS), a diversity promoting alternative to BS for approximate inference. DBS produces sequences that are significantly different from each other by incorporating diversity constraints within groups of candidate sequences during decoding; moreover, it achieves this with minimal computational or memory overhead. We demonstrate that our method improves both diversity and quality of decoded sequences over existing techniques on two visually-grounded language generation tasks---image captioning and visual question generation---particularly on complex scenes containing diverse visual content. We also show similar improvements at language-only machine translation tasks, highlighting the generality of our approach.

We present a new AI task - Embodied Question Answering(EmbodiedQA) - where an agent is spawned at a random location in a 3D environment and asked a question ('What color is the car?'). In order to answer, the agent must first intelligently navigate to explore the environment, gather necessary visual information through first-person (egocentric) vision, and then answer the question ('orange'). EmbodiedQA requires a range of AI skills - language understanding, visual recognition, active perception, goal-driven navigation, commonsense reasoning, long-term memory, and grounding language into actions. In this work, we develop a dataset of questions and answers in House3D environments [1], evaluation metrics, and a hierarchical model trained with imitation and reinforcement learning.

Many structured prediction problems (particularly in vision and language domains) are ambiguous, with multiple outputs being ‘correct’ for an input {–} e.g. there are many ways of describing an image, multiple ways of translating a sentence; however, exhaustively annotating the applicability of all possible outputs is intractable due to exponentially large output spaces (e.g. all English sentences). In practice, these problems are cast as multi-class prediction, with the likelihood of only a sparse set of annotations being maximized {–} unfortunately penalizing for placing beliefs on plausible but unannotated outputs. We make and test the following hypothesis {–} for a given input, the annotations of its neighbors may serve as an additional supervisory signal. Specifically, we propose an objective that transfers supervision from neighboring examples. We first study the properties of our developed method in a controlled toy setup before reporting results on multi-label classification and two image-grounded sequence modeling tasks {–} captioning and question generation. We evaluate using standard task-specific metrics and measures of output diversity, finding consistent improvements over standard maximum likelihood training and other baselines.

Modern Visual Question Answering (VQA) models have been shown to rely heavily on superficial correlations between question and answer words learned during training -- \eg overwhelmingly reporting the type of room as kitchen or the sport being played as tennis, irrespective of the image. Most alarmingly, this shortcoming is often not well reflected during evaluation because the same strong priors exist in test distributions; however, a VQA system that fails to ground questions in image content would likely perform poorly in real-world settings. In this work, we present a novel regularization scheme for VQA that reduces this effect. We introduce a question-only model that takes as input the question encoding from the VQA model and must leverage language biases in order to succeed. We then pose training as an adversarial game between the VQA model and this question-only adversary -- discouraging the VQA model from capturing language biases in its question encoding.Further, we leverage this question-only model to estimate the mutual information between the image and answer given the question, which we maximize explicitly to encourage visual grounding. Our approach is a model agnostic training procedure and simple to implement. We show empirically that it can improve performance significantly on a bias-sensitive split of the VQA dataset for multiple base models -- achieving state-of-the-art on this task. Further, on standard VQA tasks, our approach shows significantly less drop in accuracy compared to existing bias-reducing VQA models.

We develop the first approximate inference algorithm for 1-Best (and M-Best) decoding in bidirectional neural sequence models by extending Beam Search (BS) to reason about both forward and backward time dependencies. Beam Search (BS) is a widely used approximate inference algorithm for decoding sequences from unidirectional neural sequence models. Interestingly, approximate inference in bidirectional models remains an open problem, despite their significant advantage in modeling information from both the past and future. To enable the use of bidirectional models, we present Bidirectional Beam Search (BiBS), an efficient algorithm for approximate bidirectional inference. To evaluate our method and as an interesting problem in its own right, we introduce a novel Fill-in-the-Blank Image Captioning task which requires reasoning about both past and future sentence structure to reconstruct sensible image descriptions. We use this task as well as the Visual Madlibs dataset to demonstrate the effectiveness of our approach, consistently outperforming all baseline methods.

A number of recent works have proposed techniques for end-to-end learning of communication protocols among cooperative multi-agent populations, and have simultaneously found the emergence of grounded human-interpretable language in the protocols developed by the agents, learned without any human supervision!In this paper, using a Task & Talk reference game between two agents as a testbed, we present a sequence of 'negative' results culminating in a 'positive' one -showing that while most agent-invented languages are effective (i.e.achieve near-perfect task rewards), they are decidedly not interpretable or compositional.In essence, we find that natural language does not emerge 'naturally', despite the semblance of ease of natural-language-emergence that one may gather from recent literature.We discuss how it is possible to coax the invented languages to become more and more human-like and compositional by increasing restrictions on how two agents may communicate.

In this paper, we make a simple observation that questions about images often contain premises -objects and relationships implied by the question -and that reasoning about premises can help Visual Question Answering (VQA) models respond more intelligently to irrelevant or previously unseen questions.When presented with a question that is irrelevant to an image, state-of-the-art VQA models will still answer purely based on learned language biases, resulting in nonsensical or even misleading answers.We note that a visual question is irrelevant to an image if at least one of its premises is false (i.e.not depicted in the image).We leverage this observation to construct a dataset for Question Relevance Prediction and Explanation (QRPE) by searching for false premises.We train novel question relevance detection models and show that models that reason about premises consistently outperform models that do not.We also find that forcing standard VQA models to reason about premises during training can lead to improvements on tasks requiring compositional reasoning.

As AI continues to advance, human-AI teams are inevitable. However, progress in AI is routinely measured in isolation, without a human in the loop. It is crucial to benchmark progress in AI, not just in isolation, but also in terms of how it translates to helping humans perform certain tasks, i.e., the performance of human-AI teams. In this work, we design a cooperative game — GuessWhich — to measure human-AI team performance in the specific context of the AI being a visual conversational agent. GuessWhich involves live interaction between the human and the AI. The AI, which we call ALICE, is provided an image which is unseen by the human. Following a brief description of the image, the human questions ALICE about this secret image to identify it from a fixed pool of images. We measure performance of the human-ALICE team by the number of guesses it takes the human to correctly identify the secret image after a fixed number of dialog rounds with ALICE. We compare performance of the human-ALICE teams for two versions of ALICE. Our human studies suggest a counterintuitive trend – that while AI literature shows that one version outperforms the other when paired with an AI questioner bot, we find that this improvement in AI-AI performance does not translate to improved human-AI performance. This suggests a mismatch between benchmarking of AI in isolation and in the context of human-AI teams.

We introduce the first goal-driven training for visual question answering and dialog agents. Specifically, we pose a cooperative `image guessing' game between two agents - Q-BOT and A-BOT- who communicate in natural language dialog so that Q-BOT can select an unseen image from a lineup of images. We use deep reinforcement learning (RL) to learn the policies of these agents end-to-end - from pixels to multi-agent multi-round dialog to game reward.,,We demonstrate two experimental results.,,First, as a `sanity check' demonstration of pure RL (from scratch), we show results on a synthetic world, where the agents communicate in ungrounded vocabularies, i.e., symbols with no pre-specified meanings (X, Y, Z). We find that two bots invent their own communication protocol and start using certain symbols to ask/answer about certain visual attributes (shape/color/style). Thus, we demonstrate the emergence of grounded language and communication among `visual' dialog agents with no human supervision.,,Second, we conduct large-scale real-image experiments on the VisDial dataset [5], where we pretrain on dialog data with supervised learning (SL) and show that the RL finetuned agents significantly outperform supervised pretraining. Interestingly, the RL Q-BOT learns to ask questions that A-BOT is good at, ultimately resulting in more informative dialog and a better team.

Many practical perception systems exist within larger processes which often include interactions with users or additional components that are capable of evaluating the quality of predicted solutions. In these contexts, it is beneficial to provide these oracle mechanisms with multiple highly likely hypotheses rather than a single prediction. In this work, we pose the task of producing multiple outputs as a learning problem over an ensemble of deep networks -- introducing a novel stochastic gradient descent based approach to minimize the loss with respect to an oracle. Our method is simple to implement, agnostic to both architecture and loss function, and parameter-free. Our approach achieves lower oracle error compared to existing methods on a wide range of tasks and deep architectures. We also show qualitatively that solutions produced from our approach often provide interpretable representations of task ambiguity.

With vast quantities of imagery now available online, researchers have begun to explore whether visual patterns can be discovered automatically. Here we consider the particular domain of architecture, using huge collections of street-level imagery to find visual patterns that correspond to semantic-level architectural elements distinctive to particular time periods. We use this analysis both to date buildings, as well as to discover how functionally-similar architectural elements (e.g. windows, doors, balconies, etc.) have changed over time due to evolving styles. We validate the methods by combining a large dataset of nearly 150,000 Google Street View images from Paris with a cadastre map to infer approximate construction date for each facade. Not only could our analysis be used for dating or geo- localizing buildings based on architectural features, but it also could give architects and historians new tools for confirming known theories or even discovering new ones.

Hands appear very often in egocentric video, and their appearance and pose give important cues about what people are doing and what they are paying attention to. But existing work in hand detection has made strong assumptions that work well in only simple scenarios, such as with limited interaction with other people or in lab settings. We develop methods to locate and distinguish between hands in egocentric video using strong appearance models with Convolutional Neural Networks, and introduce a simple candidate region generation approach that outperforms existing techniques at a fraction of the computational cost. We show how these high-quality bounding boxes can be used to create accurate pixelwise hand regions, and as an application, we investigate the extent to which hand segmentation alone can distinguish between different activities. We evaluate these techniques on a new dataset of 48 first-person videos of people interacting in realistic environments, with pixel-level ground truth for over 15,000 hand instances.

Geographic location is a powerful property for organizing large-scale photo collections, but only a small fraction of online photos are geo-tagged. Most work in automatically estimating geo-tags from image content is based on comparison against models of buildings or landmarks, or on matching to large reference collections of geotagged images. These approaches work well for frequently photographed places like major cities and tourist destinations, but fail for photos taken in sparsely photographed places where few reference photos exist. Here we consider how to recognize general geo-informative attributes of a photo, e.g. the elevation gradient, population density, demographics, etc. of where it was taken, instead of trying to estimate a precise geo-tag. We learn models for these attributes using a large (noisy) set of geo-tagged images from Flickr by training deep convolutional neural networks (CNNs). We evaluate on over a dozen attributes, showing that while automatically recognizing some attributes is very difficult, others can be automatically estimated with about the same accuracy as a human.

Climate models that predict polar ice sheet behavior require accurate measurements of the bedrock-ice and ice-air boundaries in ground-penetrating radar imagery. Identifying these features is typically performed by hand, which can be tedious and error prone. We propose an approach for automatically estimating layer boundaries by viewing this task as a probabilistic inference problem. Our solution uses Markov-Chain Monte Carlo to sample from the joint distribution over all possible layers conditioned on an image. Layer boundaries can then be estimated from the expectation over this distribution, and confidence intervals can be estimated from the variance of the samples. We evaluate the method on 560 echograms collected in Antarctica, and compare to a state-of-the-art technique with respect to hand-labeled images. These experiments show an approximately 50% reduction in error for tracing both bedrock and surface layers.