We introduce LaViLa, a new approach to learning video-language representations by leveraging Large Language Models (LLMs). We repurpose pre-trained LLMs to be conditioned on visual input, and finetune them to create automatic video narrators. Our auto-generated narrations offer a number of advantages, including dense coverage of long videos, better temporal synchronization of the visual information and text, and much higher diversity of text. The video-text embedding learned contrastively with these additional auto-generated narrations outperforms the previous state-of-the-art on multiple first-person and third-person video tasks, both in zero-shot and finetuned setups. Most notably, LaViLa obtains an absolute gain of 10.1% on EGTEA classification and 5.9% Epic-Kitchens-100 multi-instance retrieval benchmarks. Furthermore, LaViLa trained with only half the narrations from the Ego4D dataset outperforms baseline models trained on the full set, and shows positive scaling behavior on increasing pre-training data and model size.
Benefiting from masked visual modeling, self-supervised video representation learning has achieved remarkable progress. However, existing methods focus on learning representations from scratch through reconstructing low-level features like raw pixel RGB values. In this paper, we propose masked video distillation (MVD), a simple yet effective two-stage masked feature modeling framework for video representation learning: firstly we pretrain an image (or video) model by recovering low-level features of masked patches, then we use the resulting features as targets for masked feature modeling. For the choice of teacher models, we observe that students taught by video teachers perform better on temporally-heavy video tasks, while image teachers transfer stronger spatial representations for spatially-heavy video tasks. Visualization analysis also indicates different teachers produce different learned patterns for students. Motivated by this observation, we design a spatial-temporal co-teaching method for MVD. Specifically, we distill student models from both video teachers and image teachers by masked feature modeling. Extensive experimental results demonstrate that video transformers pretrained with spatial-temporal co-teaching outperform models distilled with a single teacher on a multitude of video datasets. Our MVD with vanilla ViT achieves state-of-the-art performance compared with previous supervised or self-supervised methods on several challenging video downstream tasks. For example, with the ViT-Large model, our MVD achieves 86.4% and 76.7% Top-1 accuracy on Kinetics-400 and Something-Something-v2, outperforming VideoMAE by 1.2% and 2.4% respectively. When a larger ViT-Huge model is adopted, MVD achieves the state-of-the-art performance with 77.3% Top-1 accuracy on Something-Something-v2 and 41.1 mAP on AVA v2.2. Code will be available at \url{https://github.com/ruiwang2021/mvd}.
Previous work on action representation learning focused on global representations for short video clips. In contrast, many practical applications, such as video alignment, strongly demand learning the intensive representation of long videos. In this paper, we introduce a new framework of contrastive action representation learning (CARL) to learn frame-wise action representation in a self-supervised or weakly-supervised manner, especially for long videos. Specifically, we introduce a simple but effective video encoder that considers both spatial and temporal context by combining convolution and transformer. Inspired by the recent massive progress in self-supervised learning, we propose a new sequence contrast loss (SCL) applied to two related views obtained by expanding a series of spatio-temporal data in two versions. One is the self-supervised version that optimizes embedding space by minimizing KL-divergence between sequence similarity of two augmented views and prior Gaussian distribution of timestamp distance. The other is the weakly-supervised version that builds more sample pairs among videos using video-level labels by dynamic time wrapping (DTW). Experiments on FineGym, PennAction, and Pouring datasets show that our method outperforms previous state-of-the-art by a large margin for downstream fine-grained action classification and even faster inference. Surprisingly, although without training on paired videos like in previous works, our self-supervised version also shows outstanding performance in video alignment and fine-grained frame retrieval tasks.
We present a simple approach which can turn a ViT encoder into an efficient video model, which can seamlessly work with both image and video inputs. By sparsely sampling the inputs, the model is able to do training and inference from both inputs. The model is easily scalable and can be adapted to large-scale pre-trained ViTs without requiring full finetuning. The model achieves SOTA results and the code will be open-sourced.
The foundation models have recently shown excellent performance on a variety of downstream tasks in computer vision. However, most existing vision foundation models simply focus on image-level pretraining and adpation, which are limited for dynamic and complex video-level understanding tasks. To fill the gap, we present general video foundation models, InternVideo, by taking advantage of both generative and discriminative self-supervised video learning. Specifically, InternVideo efficiently explores masked video modeling and video-language contrastive learning as the pretraining objectives, and selectively coordinates video representations of these two complementary frameworks in a learnable manner to boost various video applications. Without bells and whistles, InternVideo achieves state-of-the-art performance on 39 video datasets from extensive tasks including video action recognition/detection, video-language alignment, and open-world video applications. Especially, our methods can obtain 91.1% and 77.2% top-1 accuracy on the challenging Kinetics-400 and Something-Something V2 benchmarks, respectively. All of these results effectively show the generality of our InternVideo for video understanding. The code will be released at https://github.com/OpenGVLab/InternVideo .
We propose Spatio-temporal Crop Aggregation for video representation LEarning (SCALE), a novel method that enjoys high scalability at both training and inference time. Our model builds long-range video features by learning from sets of video clip-level features extracted with a pre-trained backbone. To train the model, we propose a self-supervised objective consisting of masked clip feature prediction. We apply sparsity to both the input, by extracting a random set of video clips, and to the loss function, by only reconstructing the sparse inputs. Moreover, we use dimensionality reduction by working in the latent space of a pre-trained backbone applied to single video clips. These techniques make our method not only extremely efficient to train but also highly effective in transfer learning. We demonstrate that our video representation yields state-of-the-art performance with linear, non-linear, and KNN probing on common action classification and video understanding datasets.
Existing Temporal Action Detection (TAD) methods typically take a pre-processing step in converting an input varying-length video into a fixed-length snippet representation sequence, before temporal boundary estimation and action classification. This pre-processing step would temporally downsample the video, reducing the inference resolution and hampering the detection performance in the original temporal resolution. In essence, this is due to a temporal quantization error introduced during the resolution downsampling and recovery. This could negatively impact the TAD performance, but is largely ignored by existing methods. To address this problem, in this work we introduce a novel model-agnostic post-processing method without model redesign and retraining. Specifically, we model the start and end points of action instances with a Gaussian distribution for enabling temporal boundary inference at a sub-snippet level. We further introduce an efficient Taylor-expansion based approximation, dubbed as Gaussian Approximated Post-processing (GAP). Extensive experiments demonstrate that our GAP can consistently improve a wide variety of pre-trained off-the-shelf TAD models on the challenging ActivityNet (+0.2% -0.7% in average mAP) and THUMOS (+0.2% -0.5% in average mAP) benchmarks. Such performance gains are already significant and highly comparable to those achieved by novel model designs. Also, GAP can be integrated with model training for further performance gain. Importantly, GAP enables lower temporal resolutions for more efficient inference, facilitating low-resource applications. The code will be available in https://github.com/sauradip/GAP
Due to the rapid temporal and fine-grained nature of complex human assembly atomic actions, traditional action segmentation approaches requiring the spatial (and often temporal) down sampling of video frames often loose vital fine-grained spatial and temporal information required for accurate classification within the manufacturing domain. In order to fully utilise higher resolution video data (often collected within the manufacturing domain) and facilitate real time accurate action segmentation - required for human robot collaboration - we present a novel hand location guided high resolution feature enhanced model. We also propose a simple yet effective method of deploying offline trained action recognition models for real time action segmentation on temporally short fine-grained actions, through the use of surround sampling while training and temporally aware label cleaning at inference. We evaluate our model on a novel action segmentation dataset containing 24 (+background) atomic actions from video data of a real world robotics assembly production line. Showing both high resolution hand features as well as traditional frame wide features improve fine-grained atomic action classification, and that though temporally aware label clearing our model is capable of surpassing similar encoder/decoder methods, while allowing for real time classification.
We present XKD, a novel self-supervised framework to learn meaningful representations from unlabelled videos. XKD is trained with two pseudo objectives. First, masked data reconstruction is performed to learn modality-specific representations from audio and visual streams. Next, self-supervised cross-modal knowledge distillation is performed between the two modalities through a teacher-student setup to learn complementary information. We introduce a novel domain alignment strategy to tackle domain discrepancy between audio and visual modalities enabling effective cross-modal knowledge distillation. Additionally, to develop a general-purpose network capable of handling both audio and visual streams, modality-agnostic variants of XKD are introduced, which use the same pretrained backbone for different audio and visual tasks. Our proposed cross-modal knowledge distillation improves video action classification by $8\%$ to $14\%$ on UCF101, HMDB51, and Kinetics400. Additionally, XKD improves multimodal action classification by $5.5\%$ on Kinetics-Sound. XKD shows state-of-the-art performance in sound classification on ESC50, achieving top-1 accuracy of $96.5\%$.
In this paper, a deep learning-based model for 3D human motion generation from the text is proposed via gesture action classification and an autoregressive model. The model focuses on generating special gestures that express human thinking, such as waving and nodding. To achieve the goal, the proposed method predicts expression from the sentences using a text classification model based on a pretrained language model and generates gestures using the gate recurrent unit-based autoregressive model. Especially, we proposed the loss for the embedding space for restoring raw motions and generating intermediate motions well. Moreover, the novel data augmentation method and stop token are proposed to generate variable length motions. To evaluate the text classification model and 3D human motion generation model, a gesture action classification dataset and action-based gesture dataset are collected. With several experiments, the proposed method successfully generates perceptually natural and realistic 3D human motion from the text. Moreover, we verified the effectiveness of the proposed method using a public-available action recognition dataset to evaluate cross-dataset generalization performance.
Masked Autoencoders (MAEs) learn generalizable representations for image, text, audio, video, etc., by reconstructing masked input data from tokens of the visible data. Current MAE approaches for videos rely on random patch, tube, or frame-based masking strategies to select these tokens. This paper proposes AdaMAE, an adaptive masking strategy for MAEs that is end-to-end trainable. Our adaptive masking strategy samples visible tokens based on the semantic context using an auxiliary sampling network. This network estimates a categorical distribution over spacetime-patch tokens. The tokens that increase the expected reconstruction error are rewarded and selected as visible tokens, motivated by the policy gradient algorithm in reinforcement learning. We show that AdaMAE samples more tokens from the high spatiotemporal information regions, thereby allowing us to mask 95% of tokens, resulting in lower memory requirements and faster pre-training. We conduct ablation studies on the Something-Something v2 (SSv2) dataset to demonstrate the efficacy of our adaptive sampling approach and report state-of-the-art results of 70.0% and 81.7% in top-1 accuracy on SSv2 and Kinetics-400 action classification datasets with a ViT-Base backbone and 800 pre-training epochs.
This paper proposes a self-supervised approach to learn universal facial representations from videos, that can transfer across a variety of facial analysis tasks such as Facial Attribute Recognition (FAR), Facial Expression Recognition (FER), DeepFake Detection (DFD), and Lip Synchronization (LS). Our proposed framework, named MARLIN, is a facial video masked autoencoder, that learns highly robust and generic facial embeddings from abundantly available non-annotated web crawled facial videos. As a challenging auxiliary task, MARLIN reconstructs the spatio-temporal details of the face from the densely masked facial regions which mainly include eyes, nose, mouth, lips, and skin to capture local and global aspects that in turn help in encoding generic and transferable features. Through a variety of experiments on diverse downstream tasks, we demonstrate MARLIN to be an excellent facial video encoder as well as feature extractor, that performs consistently well across a variety of downstream tasks including FAR (1.13% gain over supervised benchmark), FER (2.64% gain over unsupervised benchmark), DFD (1.86% gain over unsupervised benchmark), LS (29.36% gain for Frechet Inception Distance), and even in low data regime. Our code and models are available at https://github.com/ControlNet/MARLIN .
We launch EVA, a vision-centric foundation model to explore the limits of visual representation at scale using only publicly accessible data. EVA is a vanilla ViT pre-trained to reconstruct the masked out image-text aligned vision features conditioned on visible image patches. Via this pretext task, we can efficiently scale up EVA to one billion parameters, and sets new records on a broad range of representative vision downstream tasks, such as image recognition, video action recognition, object detection, instance segmentation and semantic segmentation without heavy supervised training. Moreover, we observe quantitative changes in scaling EVA result in qualitative changes in transfer learning performance that are not present in other models. For instance, EVA takes a great leap in the challenging large vocabulary instance segmentation task: our model achieves almost the same state-of-the-art performance on LVISv1.0 dataset with over a thousand categories and COCO dataset with only eighty categories. Beyond a pure vision encoder, EVA can also serve as a vision-centric, multi-modal pivot to connect images and text. We find initializing the vision tower of a giant CLIP from EVA can greatly stabilize the training and outperform the training from scratch counterpart with much fewer samples and less compute, providing a new direction for scaling up and accelerating the costly training of multi-modal foundation models. To facilitate future research, we release all the code and models at https://github.com/baaivision/EVA.
Temporal Action Localization (TAL) methods typically operate on top of feature sequences from a frozen snippet encoder that is pretrained with the Trimmed Action Classification (TAC) tasks, resulting in a task discrepancy problem. While existing TAL methods mitigate this issue either by retraining the encoder with a pretext task or by end-to-end fine-tuning, they commonly require an overload of high memory and computation. In this work, we introduce Soft-Landing (SoLa) strategy, an efficient yet effective framework to bridge the transferability gap between the pretrained encoder and the downstream tasks by incorporating a light-weight neural network, i.e., a SoLa module, on top of the frozen encoder. We also propose an unsupervised training scheme for the SoLa module; it learns with inter-frame Similarity Matching that uses the frame interval as its supervisory signal, eliminating the need for temporal annotations. Experimental evaluation on various benchmarks for downstream TAL tasks shows that our method effectively alleviates the task discrepancy problem with remarkable computational efficiency.
This paper studies audio-visual noise suppression for egocentric videos -- where the speaker is not captured in the video. Instead, potential noise sources are visible on screen with the camera emulating the off-screen speaker's view of the outside world. This setting is different from prior work in audio-visual speech enhancement that relies on lip and facial visuals. In this paper, we first demonstrate that egocentric visual information is helpful for noise suppression. We compare object recognition and action classification-based visual feature extractors and investigate methods to align audio and visual representations. Then, we examine different fusion strategies for the aligned features, and locations within the noise suppression model to incorporate visual information. Experiments demonstrate that visual features are most helpful when used to generate additive correction masks. Finally, in order to ensure that the visual features are discriminative with respect to different noise types, we introduce a multi-task learning framework that jointly optimizes audio-visual noise suppression and video-based acoustic event detection. This proposed multi-task framework outperforms the audio-only baseline on all metrics, including a 0.16 PESQ improvement. Extensive ablations reveal the improved performance of the proposed model with multiple active distractors, overall noise types, and across different SNRs.
Due to the numerous potential applications in visual surveillance and nighttime driving, recognizing human action in low-light conditions remains a difficult problem in computer vision. Existing methods separate action recognition and dark enhancement into two distinct steps to accomplish this task. However, isolating the recognition and enhancement impedes end-to-end learning of the space-time representation for video action classification. This paper presents a domain adaptation-based action recognition approach that uses adversarial learning in cross-domain settings to learn cross-domain action recognition. Supervised learning can train it on a large amount of labeled data from the source domain (daytime action sequences). However, it uses deep domain invariant features to perform unsupervised learning on many unlabelled data from the target domain (night-time action sequences). The resulting augmented model, named 3D-DiNet can be trained using standard backpropagation with an additional layer. It achieves SOTA performance on InFAR and XD145 actions datasets.
The objective of this paper is an efficient training method for video tasks. We make three contributions: (1) We propose Turbo training, a simple and versatile training paradigm for Transformers on multiple video tasks. (2) We illustrate the advantages of Turbo training on action classification, video-language representation learning, and long-video activity classification, showing that Turbo training can largely maintain competitive performance while achieving almost 4X speed-up and significantly less memory consumption. (3) Turbo training enables long-schedule video-language training and end-to-end long-video training, delivering competitive or superior performance than previous works, which were infeasible to train under limited resources.
Human action recognition in computer vision has been widely studied in recent years. However, most algorithms consider only certain action specially with even high computational cost. That is not suitable for practical applications with multiple actions to be identified with low computational cost. To meet various application scenarios, this paper presents a unified human action recognition framework composed of two modules, i.e., multi-form human detection and corresponding action classification. Among them, an open-source dataset is constructed to train a multi-form human detection model that distinguishes a human being's whole body, upper body or part body, and the followed action classification model is adopted to recognize such action as falling, sleeping or on-duty, etc. Some experimental results show that the unified framework is effective for various application scenarios. It is expected to be a new application-driven AI paradigm for human action recognition.
When applied to autonomous vehicle (AV) settings, action recognition can enhance an environment model's situational awareness. This is especially prevalent in scenarios where traditional geometric descriptions and heuristics in AVs are insufficient. However, action recognition has traditionally been studied for humans, and its limited adaptability to noisy, un-clipped, un-pampered, raw RGB data has limited its application in other fields. To push for the advancement and adoption of action recognition into AVs, this work proposes a novel two-stage action recognition system, termed RALACs. RALACs formulates the problem of action recognition for road scenes, and bridges the gap between it and the established field of human action recognition. This work shows how attention layers can be useful for encoding the relations across agents, and stresses how such a scheme can be class-agnostic. Furthermore, to address the dynamic nature of agents on the road, RALACs constructs a novel approach to adapting Region of Interest (ROI) Alignment to agent tracks for downstream action classification. Finally, our scheme also considers the problem of active agent detection, and utilizes a novel application of fusing optical flow maps to discern relevant agents in a road scene. We show that our proposed scheme can outperform the baseline on the ICCV2021 Road Challenge dataset and by deploying it on a real vehicle platform, we provide preliminary insight to the usefulness of action recognition in decision making.
Distinguishing if an action is performed as intended or if an intended action fails is an important skill that not only humans have, but that is also important for intelligent systems that operate in human environments. Recognizing if an action is unintentional or anticipating if an action will fail, however, is not straightforward due to lack of annotated data. While videos of unintentional or failed actions can be found in the Internet in abundance, high annotation costs are a major bottleneck for learning networks for these tasks. In this work, we thus study the problem of self-supervised representation learning for unintentional action prediction. While previous works learn the representation based on a local temporal neighborhood, we show that the global context of a video is needed to learn a good representation for the three downstream tasks: unintentional action classification, localization and anticipation. In the supplementary material, we show that the learned representation can be used for detecting anomalies in videos as well.