This paper addresses the task of video question answering (videoQA) via a decomposed multi-stage, modular reasoning framework. Previous modular methods have shown promise with a single planning stage ungrounded in visual content. However, through a simple and effective baseline, we find that such systems can lead to brittle behavior in practice for challenging videoQA settings. Thus, unlike traditional single-stage planning methods, we propose a multi-stage system consisting of an event parser, a grounding stage, and a final reasoning stage in conjunction with an external memory. All stages are training-free, and performed using few-shot prompting of large models, creating interpretable intermediate outputs at each stage. By decomposing the underlying planning and task complexity, our method, MoReVQA, improves over prior work on standard videoQA benchmarks (NExT-QA, iVQA, EgoSchema, ActivityNet-QA) with state-of-the-art results, and extensions to related tasks (grounded videoQA, paragraph captioning).
We introduce a new attention mechanism, dubbed structural self-attention (StructSA), that leverages rich correlation patterns naturally emerging in key-query interactions of attention. StructSA generates attention maps by recognizing space-time structures of key-query correlations via convolution and uses them to dynamically aggregate local contexts of value features. This effectively leverages rich structural patterns in images and videos such as scene layouts, object motion, and inter-object relations. Using StructSA as a main building block, we develop the structural vision transformer (StructViT) and evaluate its effectiveness on both image and video classification tasks, achieving state-of-the-art results on ImageNet-1K, Kinetics-400, Something-Something V1 & V2, Diving-48, and FineGym.
Learning generalizable visual representations from Internet data has yielded promising results for robotics. Yet, prevailing approaches focus on pre-training 2D representations, being sub-optimal to deal with occlusions and accurately localize objects in complex 3D scenes. Meanwhile, 3D representation learning has been limited to single-object understanding. To address these limitations, we introduce a novel 3D pre-training framework for robotics named SUGAR that captures semantic, geometric and affordance properties of objects through 3D point clouds. We underscore the importance of cluttered scenes in 3D representation learning, and automatically construct a multi-object dataset benefiting from cost-free supervision in simulation. SUGAR employs a versatile transformer-based model to jointly address five pre-training tasks, namely cross-modal knowledge distillation for semantic learning, masked point modeling to understand geometry structures, grasping pose synthesis for object affordance, 3D instance segmentation and referring expression grounding to analyze cluttered scenes. We evaluate our learned representation on three robotic-related tasks, namely, zero-shot 3D object recognition, referring expression grounding, and language-driven robotic manipulation. Experimental results show that SUGAR's 3D representation outperforms state-of-the-art 2D and 3D representations.
An ideal model for dense video captioning -- predicting captions localized temporally in a video -- should be able to handle long input videos, predict rich, detailed textual descriptions, and be able to produce outputs before processing the entire video. Current state-of-the-art models, however, process a fixed number of downsampled frames, and make a single full prediction after seeing the whole video. We propose a streaming dense video captioning model that consists of two novel components: First, we propose a new memory module, based on clustering incoming tokens, which can handle arbitrarily long videos as the memory is of a fixed size. Second, we develop a streaming decoding algorithm that enables our model to make predictions before the entire video has been processed. Our model achieves this streaming ability, and significantly improves the state-of-the-art on three dense video captioning benchmarks: ActivityNet, YouCook2 and ViTT. Our code is released at https://github.com/google-research/scenic.
In this paper, we address web-scale visual entity recognition, specifically the task of mapping a given query image to one of the 6 million existing entities in Wikipedia. One way of approaching a problem of such scale is using dual-encoder models (eg CLIP), where all the entity names and query images are embedded into a unified space, paving the way for an approximate k-NN search. Alternatively, it is also possible to re-purpose a captioning model to directly generate the entity names for a given image. In contrast, we introduce a novel Generative Entity Recognition (GER) framework, which given an input image learns to auto-regressively decode a semantic and discriminative ``code'' identifying the target entity. Our experiments demonstrate the efficacy of this GER paradigm, showcasing state-of-the-art performance on the challenging OVEN benchmark. GER surpasses strong captioning, dual-encoder, visual matching and hierarchical classification baselines, affirming its advantage in tackling the complexities of web-scale recognition.
This paper introduces SceneCraft, a Large Language Model (LLM) Agent converting text descriptions into Blender-executable Python scripts which render complex scenes with up to a hundred 3D assets. This process requires complex spatial planning and arrangement. We tackle these challenges through a combination of advanced abstraction, strategic planning, and library learning. SceneCraft first models a scene graph as a blueprint, detailing the spatial relationships among assets in the scene. SceneCraft then writes Python scripts based on this graph, translating relationships into numerical constraints for asset layout. Next, SceneCraft leverages the perceptual strengths of vision-language foundation models like GPT-V to analyze rendered images and iteratively refine the scene. On top of this process, SceneCraft features a library learning mechanism that compiles common script functions into a reusable library, facilitating continuous self-improvement without expensive LLM parameter tuning. Our evaluation demonstrates that SceneCraft surpasses existing LLM-based agents in rendering complex scenes, as shown by its adherence to constraints and favorable human assessments. We also showcase the broader application potential of SceneCraft by reconstructing detailed 3D scenes from the Sintel movie and guiding a video generative model with generated scenes as intermediary control signal.
As foundation models become more popular, there is a growing need to efficiently finetune them for downstream tasks. Although numerous adaptation methods have been proposed, they are designed to be efficient only in terms of how many parameters are trained. They, however, typically still require backpropagating gradients throughout the model, meaning that their training-time and -memory cost does not reduce as significantly. We propose an adaptation method which does not backpropagate gradients through the backbone. We achieve this by designing a lightweight network in parallel that operates on features from the frozen, pretrained backbone. As a result, our method is efficient not only in terms of parameters, but also in training-time and memory usage. Our approach achieves state-of-the-art accuracy-parameter trade-offs on the popular VTAB benchmark, and we further show how we outperform prior works with respect to training-time and -memory usage too. We further demonstrate the training efficiency and scalability of our method by adapting a vision transformer backbone of 4 billion parameters for the computationally demanding task of video classification, without any intricate model parallelism. Here, we outperform a prior adaptor-based method which could only scale to a 1 billion parameter backbone, or fully-finetuning a smaller backbone, with the same GPU and less training time.
We present a novel unconditional video generative model designed to address long-term spatial and temporal dependencies. To capture these dependencies, our approach incorporates a hybrid explicit-implicit tri-plane representation inspired by 3D-aware generative frameworks developed for three-dimensional object representation and employs a singular latent code to model an entire video sequence. Individual video frames are then synthesized from an intermediate tri-plane representation, which itself is derived from the primary latent code. This novel strategy reduces computational complexity by a factor of $2$ as measured in FLOPs. Consequently, our approach facilitates the efficient and temporally coherent generation of videos. Moreover, our joint frame modeling approach, in contrast to autoregressive methods, mitigates the generation of visual artifacts. We further enhance the model's capabilities by integrating an optical flow-based module within our Generative Adversarial Network (GAN) based generator architecture, thereby compensating for the constraints imposed by a smaller generator size. As a result, our model is capable of synthesizing high-fidelity video clips at a resolution of $256\times256$ pixels, with durations extending to more than $5$ seconds at a frame rate of 30 fps. The efficacy and versatility of our approach are empirically validated through qualitative and quantitative assessments across three different datasets comprising both synthetic and real video clips.
Large language models have achieved great success in recent years, so as their variants in vision. Existing vision-language models can describe images in natural languages, answer visual-related questions, or perform complex reasoning about the image. However, it is yet unclear how localization tasks, such as word grounding or referring localization, can be performed using large language models. In this work, we aim to develop a vision-language model that can take locations, for example, a set of points or boxes, as either inputs or outputs. When taking locations as inputs, the model performs location-conditioned captioning, which generates captions for the indicated object or region. When generating locations as outputs, our model regresses pixel coordinates for each output word generated by the language model, and thus performs dense word grounding. Our model is pre-trained on the Localized Narrative dataset, which contains pixel-word-aligned captioning from human attention. We show our model can be applied to various location-aware vision-language tasks, including referring localization, location-conditioned captioning, and dense object captioning, archiving state-of-the-art performance on RefCOCO and Visual Genome. Project page: https://jerryxu.net/PixelLLM .
Recent approaches to point tracking are able to recover the trajectory of any scene point through a large portion of a video despite the presence of occlusions. They are, however, too slow in practice to track every point observed in a single frame in a reasonable amount of time. This paper introduces DOT, a novel, simple and efficient method for solving this problem. It first extracts a small set of tracks from key regions at motion boundaries using an off-the-shelf point tracking algorithm. Given source and target frames, DOT then computes rough initial estimates of a dense flow field and visibility mask through nearest-neighbor interpolation, before refining them using a learnable optical flow estimator that explicitly handles occlusions and can be trained on synthetic data with ground-truth correspondences. We show that DOT is significantly more accurate than current optical flow techniques, outperforms sophisticated "universal" trackers like OmniMotion, and is on par with, or better than, the best point tracking algorithms like CoTracker while being at least two orders of magnitude faster. Quantitative and qualitative experiments with synthetic and real videos validate the promise of the proposed approach. Code, data, and videos showcasing the capabilities of our approach are available in the project webpage: https://16lemoing.github.io/dot .