VideoPoet: A Large Language Model for Zero-Shot Video Generation

We present VideoPoet, a language model capable of synthesizing high-quality video, with matching audio, from a large variety of conditioning signals. VideoPoet employs a decoder-only transformer architecture that processes multimodal inputs -- including images, videos, text, and audio. The training protocol follows that of Large Language Models (LLMs), consisting of two stages: pretraining and task-specific adaptation. During pretraining, VideoPoet incorporates a mixture of multimodal generative objectives within an autoregressive Transformer framework. The pretrained LLM serves as a foundation that can be adapted for a range of video generation tasks. We present empirical results demonstrating the model's state-of-the-art capabilities in zero-shot video generation, specifically highlighting VideoPoet's ability to generate high-fidelity motions. Project page: http://sites.research.google/videopoet/

Photorealistic Video Generation with Diffusion Models

We present W.A.L.T, a transformer-based approach for photorealistic video generation via diffusion modeling. Our approach has two key design decisions. First, we use a causal encoder to jointly compress images and videos within a unified latent space, enabling training and generation across modalities. Second, for memory and training efficiency, we use a window attention architecture tailored for joint spatial and spatiotemporal generative modeling. Taken together these design decisions enable us to achieve state-of-the-art performance on established video (UCF-101 and Kinetics-600) and image (ImageNet) generation benchmarks without using classifier free guidance. Finally, we also train a cascade of three models for the task of text-to-video generation consisting of a base latent video diffusion model, and two video super-resolution diffusion models to generate videos of $512 \times 896$ resolution at $8$ frames per second.

Which way is `right'?: Uncovering limitations of Vision-and-Language Navigation model

The challenging task of Vision-and-Language Navigation (VLN) requires embodied agents to follow natural language instructions to reach a goal location or object (e.g. `walk down the hallway and turn left at the piano'). For agents to complete this task successfully, they must be able to ground objects referenced into the instruction (e.g.`piano') into the visual scene as well as ground directional phrases (e.g.`turn left') into actions. In this work we ask the following question -- to what degree are spatial and directional language cues informing the navigation model's decisions? We propose a series of simple masking experiments to inspect the model's reliance on different parts of the instruction. Surprisingly we uncover that certain top performing models rely only on the noun tokens of the instructions. We propose two training methods to alleviate this concerning limitation.

Text and Click inputs for unambiguous open vocabulary instance segmentation

Segmentation localizes objects in an image on a fine-grained per-pixel scale. Segmentation benefits by humans-in-the-loop to provide additional input of objects to segment using a combination of foreground or background clicks. Tasks include photoediting or novel dataset annotation, where human annotators leverage an existing segmentation model instead of drawing raw pixel level annotations. We propose a new segmentation process, Text + Click segmentation, where a model takes as input an image, a text phrase describing a class to segment, and a single foreground click specifying the instance to segment. Compared to previous approaches, we leverage open-vocabulary image-text models to support a wide-range of text prompts. Conditioning segmentations on text prompts improves the accuracy of segmentations on novel or unseen classes. We demonstrate that the combination of a single user-specified foreground click and a text prompt allows a model to better disambiguate overlapping or co-occurring semantic categories, such as "tie", "suit", and "person". We study these results across common segmentation datasets such as refCOCO, COCO, VOC, and OpenImages. Source code available here.

Transformer-based Localization from Embodied Dialog with Large-scale Pre-training

We address the challenging task of Localization via Embodied Dialog (LED). Given a dialog from two agents, an Observer navigating through an unknown environment and a Locator who is attempting to identify the Observer's location, the goal is to predict the Observer's final location in a map. We develop a novel LED-Bert architecture and present an effective pretraining strategy. We show that a graph-based scene representation is more effective than the top-down 2D maps used in prior works. Our approach outperforms previous baselines.

Learning a Visually Grounded Memory Assistant

We introduce a novel interface for large scale collection of human memory and assistance. Using the 3D Matterport simulator we create a realistic indoor environments in which we have people perform specific embodied memory tasks that mimic household daily activities. This interface was then deployed on Amazon Mechanical Turk allowing us to test and record human memory, navigation and needs for assistance at a large scale that was previously impossible. Using the interface we collect the `The Visually Grounded Memory Assistant Dataset' which is aimed at developing our understanding of (1) the information people encode during navigation of 3D environments and (2) conditions under which people ask for memory assistance. Additionally we experiment with with predicting when people will ask for assistance using models trained on hand-selected visual and semantic features. This provides an opportunity to build stronger ties between the machine-learning and cognitive-science communities through learned models of human perception, memory, and cognition.

No RL, No Simulation: Learning to Navigate without Navigating

Most prior methods for learning navigation policies require access to simulation environments, as they need online policy interaction and rely on ground-truth maps for rewards. However, building simulators is expensive (requires manual effort for each and every scene) and creates challenges in transferring learned policies to robotic platforms in the real-world, due to the sim-to-real domain gap. In this paper, we pose a simple question: Do we really need active interaction, ground-truth maps or even reinforcement-learning (RL) in order to solve the image-goal navigation task? We propose a self-supervised approach to learn to navigate from only passive videos of roaming. Our approach, No RL, No Simulator (NRNS), is simple and scalable, yet highly effective. NRNS outperforms RL-based formulations by a significant margin. We present NRNS as a strong baseline for any future image-based navigation tasks that use RL or Simulation.

Where Are You? Localization from Embodied Dialog

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.

Tripping through time: Efficient Localization of Activities in Videos

Localizing moments in untrimmed videos via language queries is a new and interesting task that requires the ability to accurately ground language into video. Previous works have approached this task by processing the entire video, often more than once, to localize relevant activities. In the real world applications that this task lends itself to, such as surveillance, efficiency a is pivotal trait of a system. In this paper, we present TripNet, an end-to-end system that uses a gated attention architecture to model fine-grained textual and visual representations in order to align text and video content. Furthermore, TripNet uses reinforcement learning to efficiently localize relevant activity clips in long videos, by learning how to intelligently skip around the video. In our evaluation over Charades-STA, ActivityNet Captions and the TACoS dataset, we find that TripNet achieves high accuracy and saves processing time by only looking at 32-41% of the entire video.

Action2Vec: A Crossmodal Embedding Approach to Action Learning

We describe a novel cross-modal embedding space for actions, named Action2Vec, which combines linguistic cues from class labels with spatio-temporal features derived from video clips. Our approach uses a hierarchical recurrent network to capture the temporal structure of video features. We train our embedding using a joint loss that combines classification accuracy with similarity to Word2Vec semantics. We evaluate Action2Vec by performing zero shot action recognition and obtain state of the art results on three standard datasets. In addition, we present two novel analogy tests which quantify the extent to which our joint embedding captures distributional semantics. This is the first joint embedding space to combine verbs and action videos, and the first to be thoroughly evaluated with respect to its distributional semantics.