Self-supervised representation learning approaches have recently surpassed their supervised learning counterparts on downstream tasks like object detection and image classification. Somewhat mysteriously the recent gains in performance come from training instance classification models, treating each image and it's augmented versions as samples of a single class. In this work, we first present quantitative experiments to demystify these gains. We demonstrate that approaches like MOCO and PIRL learn occlusion-invariant representations. However, they fail to capture viewpoint and category instance invariance which are crucial components for object recognition. Second, we demonstrate that these approaches obtain further gains from access to a clean object-centric training dataset like Imagenet. Finally, we propose an approach to leverage unstructured videos to learn representations that possess higher viewpoint invariance. Our results show that the learned representations outperform MOCOv2 trained on the same data in terms of invariances encoded and the performance on downstream image classification and semantic segmentation tasks.
We present an approach to infer the 3D shape, texture, and camera pose for an object from a single RGB image, using only category-level image collections with foreground masks as supervision. We represent the shape as an image-conditioned implicit function that transforms the surface of a sphere to that of the predicted mesh, while additionally predicting the corresponding texture. To derive supervisory signal for learning, we enforce that: a) our predictions when rendered should explain the available image evidence, and b) the inferred 3D structure should be geometrically consistent with learned pixel to surface mappings. We empirically show that our approach improves over prior work that leverages similar supervision, and in fact performs competitively to methods that use stronger supervision. Finally, as our method enables learning with limited supervision, we qualitatively demonstrate its applicability over a set of about 30 object categories.
In this paper, we focus on the task of extracting visual correspondences across videos. Given a query video clip from an action class, we aim to align it with training videos in space and time. Obtaining training data for such a fine-grained alignment task is challenging and often ambiguous. Hence, we propose a novel alignment procedure that learns such correspondence in space and time via cross video cycle-consistency. During training, given a pair of videos, we compute cycles that connect patches in a given frame in the first video by matching through frames in the second video. Cycles that connect overlapping patches together are encouraged to score higher than cycles that connect non-overlapping patches. Our experiments on the Penn Action and Pouring datasets demonstrate that the proposed method can successfully learn to correspond semantically similar patches across videos, and learns representations that are sensitive to object and action states.
Exploration is one of the core challenges in reinforcement learning. A common formulation of curiosity-driven exploration uses the difference between the real future and the future predicted by a learned model. However, predicting the future is an inherently difficult task which can be ill-posed in the face of stochasticity. In this paper, we introduce an alternative form of curiosity that rewards novel associations between different senses. Our approach exploits multiple modalities to provide a stronger signal for more efficient exploration. Our method is inspired by the fact that, for humans, both sight and sound play a critical role in exploration. We present results on several Atari environments and Habitat (a photorealistic navigation simulator), showing the benefits of using an audio-visual association model for intrinsically guiding learning agents in the absence of external rewards. For videos and code, see https://vdean.github.io/audio-curiosity.html.
Truly intelligent agents need to capture the interplay of all their senses to build a rich physical understanding of their world. In robotics, we have seen tremendous progress in using visual and tactile perception; however, we have often ignored a key sense: sound. This is primarily due to the lack of data that captures the interplay of action and sound. In this work, we perform the first large-scale study of the interactions between sound and robotic action. To do this, we create the largest available sound-action-vision dataset with 15,000 interactions on 60 objects using our robotic platform Tilt-Bot. By tilting objects and allowing them to crash into the walls of a robotic tray, we collect rich four-channel audio information. Using this data, we explore the synergies between sound and action and present three key insights. First, sound is indicative of fine-grained object class information, e.g., sound can differentiate a metal screwdriver from a metal wrench. Second, sound also contains information about the causal effects of an action, i.e. given the sound produced, we can predict what action was applied to the object. Finally, object representations derived from audio embeddings are indicative of implicit physical properties. We demonstrate that on previously unseen objects, audio embeddings generated through interactions can predict forward models 24% better than passive visual embeddings. Project videos and data are at https://dhiraj100892.github.io/swoosh/
This work studies the problem of object goal navigation which involves navigating to an instance of the given object category in unseen environments. End-to-end learning-based navigation methods struggle at this task as they are ineffective at exploration and long-term planning. We propose a modular system called, `Goal-Oriented Semantic Exploration' which builds an episodic semantic map and uses it to explore the environment efficiently based on the goal object category. Empirical results in visually realistic simulation environments show that the proposed model outperforms a wide range of baselines including end-to-end learning-based methods as well as modular map-based methods and led to the winning entry of the CVPR-2020 Habitat ObjectNav Challenge. Ablation analysis indicates that the proposed model learns semantic priors of the relative arrangement of objects in a scene, and uses them to explore efficiently. Domain-agnostic module design allow us to transfer our model to a mobile robot platform and achieve similar performance for object goal navigation in the real-world.
In this paper, we address the discovery of robotic options from demonstrations in an unsupervised manner. Specifically, we present a framework to jointly learn low-level control policies and higher-level policies of how to use them from demonstrations of a robot performing various tasks. By representing options as continuous latent variables, we frame the problem of learning these options as latent variable inference. We then present a temporal formulation of variational inference based on a temporal factorization of trajectory likelihoods,that allows us to infer options in an unsupervised manner. We demonstrate the ability of our framework to learn such options across three robotic demonstration datasets.
This paper formulates hypothesis verification as an RL problem. Specifically, we aim to build an agent that, given a hypothesis about the dynamics of the world, can take actions to generate observations which can help predict whether the hypothesis is true or false. Existing RL algorithms fail to solve this task, even for simple environments. In order to train the agents, we exploit the underlying structure of many hypotheses, factorizing them as {pre-condition, action sequence, post-condition} triplets. By leveraging this structure we show that RL agents are able to succeed at the task. Furthermore, subsequent fine-tuning of the policies allows the agent to correctly verify hypotheses not amenable to the above factorization.
In this paper, we study the task of embodied interactive learning for object detection. Given a set of environments (and some labeling budget), our goal is to learn an object detector by having an agent select what data to obtain labels for. How should an exploration policy decide which trajectory should be labeled? One possibility is to use a trained object detector's failure cases as an external reward. However, this will require labeling millions of frames required for training RL policies, which is infeasible. Instead, we explore a self-supervised approach for training our exploration policy by introducing a notion of semantic curiosity. Our semantic curiosity policy is based on a simple observation -- the detection outputs should be consistent. Therefore, our semantic curiosity rewards trajectories with inconsistent labeling behavior and encourages the exploration policy to explore such areas. The exploration policy trained via semantic curiosity generalizes to novel scenes and helps train an object detector that outperforms baselines trained with other possible alternatives such as random exploration, prediction-error curiosity, and coverage-maximizing exploration.