We present an approach for building an active agent that learns to segment its visual observations into individual objects by interacting with its environment in a completely self-supervised manner. The agent uses its current segmentation model to infer pixels that constitute objects and refines the segmentation model by interacting with these pixels. The model learned from over 50K interactions generalizes to novel objects and backgrounds. To deal with noisy training signal for segmenting objects obtained by self-supervised interactions, we propose robust set loss. A dataset of robot's interactions along-with a few human labeled examples is provided as a benchmark for future research. We test the utility of the learned segmentation model by providing results on a downstream vision-based control task of rearranging multiple objects into target configurations from visual inputs alone. Videos, code, and robotic interaction dataset are available at https://pathak22.github.io/seg-by-interaction/
This paper studies the problem of detecting acute intracranial hemorrhage on head computed tomography (CT) scans. We formulate it as a pixel-wise labeling task of the frames that constitute a single head scan. The standard approach for this task is the fully convolutional network (FCN) which runs on the whole image at both training and test time. We propose a patch-based approach that controls the amount of context available to the FCN, based on the observation that when radiologists are interpreting CT scans, their judgment depends primarily on local cues and does not require the whole image context. To develop and validate the system, we collected a pixel-wise labeled dataset of 591 scans that covers a wide range of hemorrhage types and imaging conditions in the real world. We show that no pretraining from natural images is needed. By aggregating the pixel-wise labeling, our system is able to make region-level, frame-level, and stack-level decisions. Our final system approaches an expert radiologist performance with a high average precision (AP) of 96.5 +/- 1.3 % for hemorrhage classification on stack level while running at 23ms per frame
This paper introduces a video dataset of spatio-temporally localized Atomic Visual Actions (AVA). The AVA dataset densely annotates 80 atomic visual actions in 430 15-minute video clips, where actions are localized in space and time, resulting in 1.58M action labels with multiple labels per person occurring frequently. The key characteristics of our dataset are: (1) the definition of atomic visual actions, rather than composite actions; (2) precise spatio-temporal annotations with possibly multiple annotations for each person; (3) exhaustive annotation of these atomic actions over 15-minute video clips; (4) people temporally linked across consecutive segments; and (5) using movies to gather a varied set of action representations. This departs from existing datasets for spatio-temporal action recognition, which typically provide sparse annotations for composite actions in short video clips. We will release the dataset publicly. AVA, with its realistic scene and action complexity, exposes the intrinsic difficulty of action recognition. To benchmark this, we present a novel approach for action localization that builds upon the current state-of-the-art methods, and demonstrates better performance on JHMDB and UCF101-24 categories. While setting a new state of the art on existing datasets, the overall results on AVA are low at 15.6% mAP, underscoring the need for developing new approaches for video understanding.
The goal of this paper is to take a single 2D image of a scene and recover the 3D structure in terms of a small set of factors: a layout representing the enclosing surfaces as well as a set of objects represented in terms of shape and pose. We propose a convolutional neural network-based approach to predict this representation and benchmark it on a large dataset of indoor scenes. Our experiments evaluate a number of practical design questions, demonstrate that we can infer this representation, and quantitatively and qualitatively demonstrate its merits compared to alternate representations.
We present a framework for learning single-view shape and pose prediction without using direct supervision for either. Our approach allows leveraging multi-view observations from unknown poses as supervisory signal during training. Our proposed training setup enforces geometric consistency between the independently predicted shape and pose from two views of the same instance. We consequently learn to predict shape in an emergent canonical (view-agnostic) frame along with a corresponding pose predictor. We show empirical and qualitative results using the ShapeNet dataset and observe encouragingly competitive performance to previous techniques which rely on stronger forms of supervision. We also demonstrate the applicability of our framework in a realistic setting which is beyond the scope of existing techniques: using a training dataset comprised of online product images where the underlying shape and pose are unknown.
The current dominant paradigm for imitation learning relies on strong supervision of expert actions to learn both 'what' and 'how' to imitate. We pursue an alternative paradigm wherein an agent first explores the world without any expert supervision and then distills its experience into a goal-conditioned skill policy with a novel forward consistency loss. In our framework, the role of the expert is only to communicate the goals (i.e., what to imitate) during inference. The learned policy is then employed to mimic the expert (i.e., how to imitate) after seeing just a sequence of images demonstrating the desired task. Our method is 'zero-shot' in the sense that the agent never has access to expert actions during training or for the task demonstration at inference. We evaluate our zero-shot imitator in two real-world settings: complex rope manipulation with a Baxter robot and navigation in previously unseen office environments with a TurtleBot. Through further experiments in VizDoom simulation, we provide evidence that better mechanisms for exploration lead to learning a more capable policy which in turn improves end task performance. Videos, models, and more details are available at https://pathak22.github.io/zeroshot-imitation/
Do visual tasks have a relationship, or are they unrelated? For instance, could having surface normals simplify estimating the depth of an image? Intuition answers these questions positively, implying existence of a structure among visual tasks. Knowing this structure has notable values; it is the concept underlying transfer learning and provides a principled way for identifying redundancies across tasks, e.g., to seamlessly reuse supervision among related tasks or solve many tasks in one system without piling up the complexity. We proposes a fully computational approach for modeling the structure of space of visual tasks. This is done via finding (first and higher-order) transfer learning dependencies across a dictionary of twenty six 2D, 2.5D, 3D, and semantic tasks in a latent space. The product is a computational taxonomic map for task transfer learning. We study the consequences of this structure, e.g. nontrivial emerged relationships, and exploit them to reduce the demand for labeled data. For example, we show that the total number of labeled datapoints needed for solving a set of 10 tasks can be reduced by roughly 2/3 (compared to training independently) while keeping the performance nearly the same. We provide a set of tools for computing and probing this taxonomical structure including a solver that users can employ to devise efficient supervision policies for their use cases.
This works presents a formulation for visual navigation that unifies map based spatial reasoning and path planning, with landmark based robust plan execution in noisy environments. Our proposed formulation is learned from data and is thus able to leverage statistical regularities of the world. This allows it to efficiently navigate in novel environments given only a sparse set of registered images as input for building representations for space. Our formulation is based on three key ideas: a learned path planner that outputs path plans to reach the goal, a feature synthesis engine that predicts features for locations along the planned path, and a learned goal-driven closed loop controller that can follow plans given these synthesized features. We test our approach for goal-driven navigation in simulated real world environments and report performance gains over competitive baseline approaches.
A major stumbling block to progress in understanding basic human interactions, such as getting out of bed or opening a refrigerator, is lack of good training data. Most past efforts have gathered this data explicitly: starting with a laundry list of action labels, and then querying search engines for videos tagged with each label. In this work, we do the reverse and search implicitly: we start with a large collection of interaction-rich video data and then annotate and analyze it. We use Internet Lifestyle Vlogs as the source of surprisingly large and diverse interaction data. We show that by collecting the data first, we are able to achieve greater scale and far greater diversity in terms of actions and actors. Additionally, our data exposes biases built into common explicitly gathered data. We make sense of our data by analyzing the central component of interaction -- hands. We benchmark two tasks: identifying semantic object contact at the video level and non-semantic contact state at the frame level. We additionally demonstrate future prediction of hands.
Learning to Optimize is a recently proposed framework for learning optimization algorithms using reinforcement learning. In this paper, we explore learning an optimization algorithm for training shallow neural nets. Such high-dimensional stochastic optimization problems present interesting challenges for existing reinforcement learning algorithms. We develop an extension that is suited to learning optimization algorithms in this setting and demonstrate that the learned optimization algorithm consistently outperforms other known optimization algorithms even on unseen tasks and is robust to changes in stochasticity of gradients and the neural net architecture. More specifically, we show that an optimization algorithm trained with the proposed method on the problem of training a neural net on MNIST generalizes to the problems of training neural nets on the Toronto Faces Dataset, CIFAR-10 and CIFAR-100.