Streaming video recognition reasons about objects and their actions in every frame of a video. A good streaming recognition model captures both long-term dynamics and short-term changes of video. Unfortunately, in most existing methods, the computational complexity grows linearly or quadratically with the length of the considered dynamics. This issue is particularly pronounced in transformer-based architectures. To address this issue, we reformulate the cross-attention in a video transformer through the lens of kernel and apply two kinds of temporal smoothing kernel: A box kernel or a Laplace kernel. The resulting streaming attention reuses much of the computation from frame to frame, and only requires a constant time update each frame. Based on this idea, we build TeSTra, a Temporal Smoothing Transformer, that takes in arbitrarily long inputs with constant caching and computing overhead. Specifically, it runs $6\times$ faster than equivalent sliding-window based transformers with 2,048 frames in a streaming setting. Furthermore, thanks to the increased temporal span, TeSTra achieves state-of-the-art results on THUMOS'14 and EPIC-Kitchen-100, two standard online action detection and action anticipation datasets. A real-time version of TeSTra outperforms all but one prior approaches on the THUMOS'14 dataset.
相關內容
Inspired by recent advances in vision transformers for object detection, we propose Li3DeTr, an end-to-end LiDAR based 3D Detection Transformer for autonomous driving, that inputs LiDAR point clouds and regresses 3D bounding boxes. The LiDAR local and global features are encoded using sparse convolution and multi-scale deformable attention respectively. In the decoder head, firstly, in the novel Li3DeTr cross-attention block, we link the LiDAR global features to 3D predictions leveraging the sparse set of object queries learnt from the data. Secondly, the object query interactions are formulated using multi-head self-attention. Finally, the decoder layer is repeated $L_{dec}$ number of times to refine the object queries. Inspired by DETR, we employ set-to-set loss to train the Li3DeTr network. Without bells and whistles, the Li3DeTr network achieves 61.3% mAP and 67.6% NDS surpassing the state-of-the-art methods with non-maximum suppression (NMS) on the nuScenes dataset and it also achieves competitive performance on the KITTI dataset. We also employ knowledge distillation (KD) using a teacher and student model that slightly improves the performance of our network.
Long samples of text from neural language models can be of poor quality. Truncation sampling algorithms--like top-$p$ or top-$k$ -- address this by setting some words' probabilities to zero at each step. This work provides framing for the aim of truncation, and an improved algorithm for that aim. We propose thinking of a neural language model as a mixture of a true distribution and a smoothing distribution that avoids infinite perplexity. In this light, truncation algorithms aim to perform desmoothing, estimating a subset of the support of the true distribution. Finding a good subset is crucial: we show that top-$p$ unnecessarily truncates high-probability words, for example causing it to truncate all words but Trump for a document that starts with Donald. We introduce $\eta$-sampling, which truncates words below an entropy-dependent probability threshold. Compared to previous algorithms, $\eta$-sampling generates more plausible long English documents according to humans, is better at breaking out of repetition, and behaves more reasonably on a battery of test distributions.
We present an explicit-grid based method for efficiently reconstructing streaming radiance fields for novel view synthesis of real world dynamic scenes. Instead of training a single model that combines all the frames, we formulate the dynamic modeling problem with an incremental learning paradigm in which per-frame model difference is trained to complement the adaption of a base model on the current frame. By exploiting the simple yet effective tuning strategy with narrow bands, the proposed method realizes a feasible framework for handling video sequences on-the-fly with high training efficiency. The storage overhead induced by using explicit grid representations can be significantly reduced through the use of model difference based compression. We also introduce an efficient strategy to further accelerate model optimization for each frame. Experiments on challenging video sequences demonstrate that our approach is capable of achieving a training speed of 15 seconds per-frame with competitive rendering quality, which attains $1000 \times$ speedup over the state-of-the-art implicit methods. Code is available at //github.com/AlgoHunt/StreamRF.
Realistic reconstruction of two hands interacting with objects is a new and challenging problem that is essential for building personalized Virtual and Augmented Reality environments. Graph Convolutional networks (GCNs) allow for the preservation of the topologies of hands poses and shapes by modeling them as a graph. In this work, we propose the THOR-Net which combines the power of GCNs, Transformer, and self-supervision to realistically reconstruct two hands and an object from a single RGB image. Our network comprises two stages; namely the features extraction stage and the reconstruction stage. In the features extraction stage, a Keypoint RCNN is used to extract 2D poses, features maps, heatmaps, and bounding boxes from a monocular RGB image. Thereafter, this 2D information is modeled as two graphs and passed to the two branches of the reconstruction stage. The shape reconstruction branch estimates meshes of two hands and an object using our novel coarse-to-fine GraFormer shape network. The 3D poses of the hands and objects are reconstructed by the other branch using a GraFormer network. Finally, a self-supervised photometric loss is used to directly regress the realistic textured of each vertex in the hands' meshes. Our approach achieves State-of-the-art results in Hand shape estimation on the HO-3D dataset (10.0mm) exceeding ArtiBoost (10.8mm). It also surpasses other methods in hand pose estimation on the challenging two hands and object (H2O) dataset by 5mm on the left-hand pose and 1 mm on the right-hand pose.
Recent research advances in salient object detection (SOD) could largely be attributed to ever-stronger multi-scale feature representation empowered by the deep learning technologies. The existing SOD deep models extract multi-scale features via the off-the-shelf encoders and combine them smartly via various delicate decoders. However, the kernel sizes in this commonly-used thread are usually "fixed". In our new experiments, we have observed that kernels of small size are preferable in scenarios containing tiny salient objects. In contrast, large kernel sizes could perform better for images with large salient objects. Inspired by this observation, we advocate the "dynamic" scale routing (as a brand-new idea) in this paper. It will result in a generic plug-in that could directly fit the existing feature backbone. This paper's key technical innovations are two-fold. First, instead of using the vanilla convolution with fixed kernel sizes for the encoder design, we propose the dynamic pyramid convolution (DPConv), which dynamically selects the best-suited kernel sizes w.r.t. the given input. Second, we provide a self-adaptive bidirectional decoder design to accommodate the DPConv-based encoder best. The most significant highlight is its capability of routing between feature scales and their dynamic collection, making the inference process scale-aware. As a result, this paper continues to enhance the current SOTA performance. Both the code and dataset are publicly available at //github.com/wuzhenyubuaa/DPNet.
We introduce a Transformer based 6D Object Pose Estimation framework VideoPose, comprising an end-to-end attention based modelling architecture, that attends to previous frames in order to estimate accurate 6D Object Poses in videos. Our approach leverages the temporal information from a video sequence for pose refinement, along with being computationally efficient and robust. Compared to existing methods, our architecture is able to capture and reason from long-range dependencies efficiently, thus iteratively refining over video sequences. Experimental evaluation on the YCB-Video dataset shows that our approach is on par with the state-of-the-art Transformer methods, and performs significantly better relative to CNN based approaches. Further, with a speed of 33 fps, it is also more efficient and therefore applicable to a variety of applications that require real-time object pose estimation. Training code and pretrained models are available at //github.com/ApoorvaBeedu/VideoPose
Convolutional neural networks have made significant progresses in edge detection by progressively exploring the context and semantic features. However, local details are gradually suppressed with the enlarging of receptive fields. Recently, vision transformer has shown excellent capability in capturing long-range dependencies. Inspired by this, we propose a novel transformer-based edge detector, \emph{Edge Detection TransformER (EDTER)}, to extract clear and crisp object boundaries and meaningful edges by exploiting the full image context information and detailed local cues simultaneously. EDTER works in two stages. In Stage I, a global transformer encoder is used to capture long-range global context on coarse-grained image patches. Then in Stage II, a local transformer encoder works on fine-grained patches to excavate the short-range local cues. Each transformer encoder is followed by an elaborately designed Bi-directional Multi-Level Aggregation decoder to achieve high-resolution features. Finally, the global context and local cues are combined by a Feature Fusion Module and fed into a decision head for edge prediction. Extensive experiments on BSDS500, NYUDv2, and Multicue demonstrate the superiority of EDTER in comparison with state-of-the-arts.
The considerable significance of Anomaly Detection (AD) problem has recently drawn the attention of many researchers. Consequently, the number of proposed methods in this research field has been increased steadily. AD strongly correlates with the important computer vision and image processing tasks such as image/video anomaly, irregularity and sudden event detection. More recently, Deep Neural Networks (DNNs) offer a high performance set of solutions, but at the expense of a heavy computational cost. However, there is a noticeable gap between the previously proposed methods and an applicable real-word approach. Regarding the raised concerns about AD as an ongoing challenging problem, notably in images and videos, the time has come to argue over the pitfalls and prospects of methods have attempted to deal with visual AD tasks. Hereupon, in this survey we intend to conduct an in-depth investigation into the images/videos deep learning based AD methods. We also discuss current challenges and future research directions thoroughly.
This paper introduces an online model for object detection in videos designed to run in real-time on low-powered mobile and embedded devices. Our approach combines fast single-image object detection with convolutional long short term memory (LSTM) layers to create an interweaved recurrent-convolutional architecture. Additionally, we propose an efficient Bottleneck-LSTM layer that significantly reduces computational cost compared to regular LSTMs. Our network achieves temporal awareness by using Bottleneck-LSTMs to refine and propagate feature maps across frames. This approach is substantially faster than existing detection methods in video, outperforming the fastest single-frame models in model size and computational cost while attaining accuracy comparable to much more expensive single-frame models on the Imagenet VID 2015 dataset. Our model reaches a real-time inference speed of up to 15 FPS on a mobile CPU.
We consider the task of weakly supervised one-shot detection. In this task, we attempt to perform a detection task over a set of unseen classes, when training only using weak binary labels that indicate the existence of a class instance in a given example. The model is conditioned on a single exemplar of an unseen class and a target example that may or may not contain an instance of the same class as the exemplar. A similarity map is computed by using a Siamese neural network to map the exemplar and regions of the target example to a latent representation space and then computing cosine similarity scores between representations. An attention mechanism weights different regions in the target example, and enables learning of the one-shot detection task using the weaker labels alone. The model can be applied to detection tasks from different domains, including computer vision object detection. We evaluate our attention Siamese networks on a one-shot detection task from the audio domain, where it detects audio keywords in spoken utterances. Our model considerably outperforms a baseline approach and yields a 42.6% average precision for detection across 10 unseen classes. Moreover, architectural developments from computer vision object detection models such as a region proposal network can be incorporated into the model architecture, and results show that performance is expected to improve by doing so.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191