In this paper, we propose a turbo receiver for joint activity detection and data decoding in grant-free massive random access, which iterates between a detector and a belief propagation (BP)-based channel decoder. Specifically, responsible for user activity detection, channel estimation, and soft data symbol detection, the detector is developed based on a bilinear inference problem that exploits the common sparsity pattern in the received pilot and data signals. The bilinear generalized approximate message passing (BiG-AMP) algorithm is adopted to solve the problem using probabilities of the transmitted data symbols estimated by the channel decoder as prior knowledge. In addition, extrinsic information is derived from the detector to improve the channel decoding accuracy of the decoder. Simulation results show significant improvements achieved by the proposed turbo receiver compared with conventional designs.
Autonomous vehicles should be able to predict the future states of its environment and respond appropriately. Specifically, predicting the behavior of surrounding human drivers is vital for such platforms to share the same road with humans. Behavior of each of the surrounding vehicles is governed by the motion of its neighbor vehicles. This paper focuses on predicting the behavior of the surrounding vehicles of an autonomous vehicle on highways. We are motivated by improving the prediction accuracy when a surrounding vehicle performs lane change and highway merging maneuvers. We propose a novel pooling strategy to capture the inter-dependencies between the neighbor vehicles. Depending solely on Euclidean trajectory representation, the existing pooling strategies do not model the context information of the maneuvers intended by a surrounding vehicle. In contrast, our pooling mechanism employs polar trajectory representation, vehicles orientation and radial velocity. This results in an implicitly maneuver-aware pooling operation. We incorporated the proposed pooling mechanism into a generative encoder-decoder model, and evaluated our method on the public NGSIM dataset. The results of maneuver-based trajectory predictions demonstrate the effectiveness of the proposed method compared with the state-of-the-art approaches. Our "Pooling Toolbox" code is available at https://github.com/m-hasan-n/pooling.
Finding a reduction of complex, high-dimensional dynamics to its essential, low-dimensional "heart" remains a challenging yet necessary prerequisite for designing efficient numerical approaches. Machine learning methods have the potential to provide a general framework to automatically discover such representations. In this paper, we consider multiscale stochastic systems with local slow-fast time scale separation and propose a new method to encode in an artificial neural network a map that extracts the slow representation from the system. The architecture of the network consists of an encoder-decoder pair that we train in a supervised manner to learn the appropriate low-dimensional embedding in the bottleneck layer. We test the method on a number of examples that illustrate the ability to discover a correct slow representation. Moreover, we provide an error measure to assess the quality of the embedding and demonstrate that pruning the network can pinpoint an essential coordinates of the system to build the slow representation.
Inpainting arbitrary missing regions is challenging because learning valid features for various masked regions is nontrivial. Though U-shaped encoder-decoder frameworks have been witnessed to be successful, most of them share a common drawback of mask unawareness in feature extraction because all convolution windows (or regions), including those with various shapes of missing pixels, are treated equally and filtered with fixed learned kernels. To this end, we propose our novel mask-aware inpainting solution. Firstly, a Mask-Aware Dynamic Filtering (MADF) module is designed to effectively learn multi-scale features for missing regions in the encoding phase. Specifically, filters for each convolution window are generated from features of the corresponding region of the mask. The second fold of mask awareness is achieved by adopting Point-wise Normalization (PN) in our decoding phase, considering that statistical natures of features at masked points differentiate from those of unmasked points. The proposed PN can tackle this issue by dynamically assigning point-wise scaling factor and bias. Lastly, our model is designed to be an end-to-end cascaded refinement one. Supervision information such as reconstruction loss, perceptual loss and total variation loss is incrementally leveraged to boost the inpainting results from coarse to fine. Effectiveness of the proposed framework is validated both quantitatively and qualitatively via extensive experiments on three public datasets including Places2, CelebA and Paris StreetView.
Recovering the 3D geometric structure of a face from a single input image is a challenging active research area in computer vision. In this paper, we present a novel method for reconstructing 3D heads from a single or multiple image(s) using a hybrid approach based on deep learning and geometric techniques. We propose an encoder-decoder network based on the U-net architecture and trained on synthetic data only. It predicts both pixel-wise normal vectors and landmarks maps from a single input photo. Landmarks are used for the pose computation and the initialization of the optimization problem, which, in turn, reconstructs the 3D head geometry by using a parametric morphable model and normal vector fields. State-of-the-art results are achieved through qualitative and quantitative evaluation tests on both single and multi-view settings. Despite the fact that the model was trained only on synthetic data, it successfully recovers 3D geometry and precise poses for real-world images.
Human-Object Interaction (HOI) detection is a task of identifying "a set of interactions" in an image, which involves the i) localization of the subject (i.e., humans) and target (i.e., objects) of interaction, and ii) the classification of the interaction labels. Most existing methods have indirectly addressed this task by detecting human and object instances and individually inferring every pair of the detected instances. In this paper, we present a novel framework, referred to by HOTR, which directly predicts a set of <human, object, interaction> triplets from an image based on a transformer encoder-decoder architecture. Through the set prediction, our method effectively exploits the inherent semantic relationships in an image and does not require time-consuming post-processing which is the main bottleneck of existing methods. Our proposed algorithm achieves the state-of-the-art performance in two HOI detection benchmarks with an inference time under 1 ms after object detection.
Digital watermarking is widely used for copyright protection. Traditional 3D watermarking approaches or commercial software are typically designed to embed messages into 3D meshes, and later retrieve the messages directly from distorted/undistorted watermarked 3D meshes. However, in many cases, users only have access to rendered 2D images instead of 3D meshes. Unfortunately, retrieving messages from 2D renderings of 3D meshes is still challenging and underexplored. We introduce a novel end-to-end learning framework to solve this problem through: 1) an encoder to covertly embed messages in both mesh geometry and textures; 2) a differentiable renderer to render watermarked 3D objects from different camera angles and under varied lighting conditions; 3) a decoder to recover the messages from 2D rendered images. From our experiments, we show that our model can learn to embed information visually imperceptible to humans, and to retrieve the embedded information from 2D renderings that undergo 3D distortions. In addition, we demonstrate that our method can also work with other renderers, such as ray tracers and real-time renderers with and without fine-tuning.
We present SrvfNet, a generative deep learning framework for the joint multiple alignment of large collections of functional data comprising square-root velocity functions (SRVF) to their templates. Our proposed framework is fully unsupervised and is capable of aligning to a predefined template as well as jointly predicting an optimal template from data while simultaneously achieving alignment. Our network is constructed as a generative encoder-decoder architecture comprising fully-connected layers capable of producing a distribution space of the warping functions. We demonstrate the strength of our framework by validating it on synthetic data as well as diffusion profiles from magnetic resonance imaging (MRI) data.
BiSeNet has been proved to be a popular two-stream network for real-time segmentation. However, its principle of adding an extra path to encode spatial information is time-consuming, and the backbones borrowed from pretrained tasks, e.g., image classification, may be inefficient for image segmentation due to the deficiency of task-specific design. To handle these problems, we propose a novel and efficient structure named Short-Term Dense Concatenate network (STDC network) by removing structure redundancy. Specifically, we gradually reduce the dimension of feature maps and use the aggregation of them for image representation, which forms the basic module of STDC network. In the decoder, we propose a Detail Aggregation module by integrating the learning of spatial information into low-level layers in single-stream manner. Finally, the low-level features and deep features are fused to predict the final segmentation results. Extensive experiments on Cityscapes and CamVid dataset demonstrate the effectiveness of our method by achieving promising trade-off between segmentation accuracy and inference speed. On Cityscapes, we achieve 71.9% mIoU on the test set with a speed of 250.4 FPS on NVIDIA GTX 1080Ti, which is 45.2% faster than the latest methods, and achieve 76.8% mIoU with 97.0 FPS while inferring on higher resolution images.
Underwater images suffer from color casts and low contrast due to wavelength- and distance-dependent attenuation and scattering. To solve these two degradation issues, we present an underwater image enhancement network via medium transmission-guided multi-color space embedding, called Ucolor. Concretely, we first propose a multi-color space encoder network, which enriches the diversity of feature representations by incorporating the characteristics of different color spaces into a unified structure. Coupled with an attention mechanism, the most discriminative features extracted from multiple color spaces are adaptively integrated and highlighted. Inspired by underwater imaging physical models, we design a medium transmission (indicating the percentage of the scene radiance reaching the camera)-guided decoder network to enhance the response of the network towards quality-degraded regions. As a result, our network can effectively improve the visual quality of underwater images by exploiting multiple color spaces embedding and the advantages of both physical model-based and learning-based methods. Extensive experiments demonstrate that our Ucolor achieves superior performance against state-of-the-art methods in terms of both visual quality and quantitative metrics.
Document grounded generation is the task of using the information provided in a document to improve text generation. This work focuses on two different document grounded generation tasks: Wikipedia Update Generation task and Dialogue response generation. Our work introduces two novel adaptations of large scale pre-trained encoder-decoder models focusing on building context driven representation of the document and enabling specific attention to the information in the document. Additionally, we provide a stronger BART baseline for these tasks. Our proposed techniques outperform existing methods on both automated (at least 48% increase in BLEU-4 points) and human evaluation for closeness to reference and relevance to the document. Furthermore, we perform comprehensive manual inspection of the generated output and categorize errors to provide insights into future directions in modeling these tasks.
Content-based music information retrieval has seen rapid progress with the adoption of deep learning. Current approaches to high-level music description typically make use of classification models, such as in auto-tagging or genre and mood classification. In this work, we propose to address music description via audio captioning, defined as the task of generating a natural language description of music audio content in a human-like manner. To this end, we present the first music audio captioning model, MusCaps, consisting of an encoder-decoder with temporal attention. Our method combines convolutional and recurrent neural network architectures to jointly process audio-text inputs through a multimodal encoder and leverages pre-training on audio data to obtain representations that effectively capture and summarise musical features in the input. Evaluation of the generated captions through automatic metrics shows that our method outperforms a baseline designed for non-music audio captioning. Through an ablation study, we unveil that this performance boost can be mainly attributed to pre-training of the audio encoder, while other design choices - modality fusion, decoding strategy and the use of attention - contribute only marginally. Our model represents a shift away from classification-based music description and combines tasks requiring both auditory and linguistic understanding to bridge the semantic gap in music information retrieval.
The fully convolutional network (FCN) with an encoder-decoder architecture has been the standard paradigm for semantic segmentation. The encoder-decoder architecture utilizes an encoder to capture multilevel feature maps, which are incorporated into the final prediction by a decoder. As the context is crucial for precise segmentation, tremendous effort has been made to extract such information in an intelligent fashion, including employing dilated/atrous convolutions or inserting attention modules. However, these endeavors are all based on the FCN architecture with ResNet or other backbones, which cannot fully exploit the context from the theoretical concept. By contrast, we introduce the Swin Transformer as the backbone to extract the context information and design a novel decoder of densely connected feature aggregation module (DCFAM) to restore the resolution and produce the segmentation map. The experimental results on two remotely sensed semantic segmentation datasets demonstrate the effectiveness of the proposed scheme.Code is available at https://github.com/WangLibo1995/GeoSeg
We present a new multi-stream 3D mesh reconstruction network (MSMR-Net) for hand pose estimation from a single RGB image. Our model consists of an image encoder followed by a mesh-convolution decoder composed of connected graph convolution layers. In contrast to previous models that form a single mesh decoding path, our decoder network incorporates multiple cross-resolution trajectories that are executed in parallel. Thus, global and local information are shared to form rich decoding representations at minor additional parameter cost compared to the single trajectory network. We demonstrate the effectiveness of our method in hand-hand and hand-object interaction scenarios at various levels of interaction. To evaluate the former scenario, we propose a method to generate RGB images of closely interacting hands. Moreoever, we suggest a metric to quantify the degree of interaction and show that close hand interactions are particularly challenging. Experimental results show that the MSMR-Net outperforms existing algorithms on the hand-object FreiHAND dataset as well as on our own hand-hand dataset.
Existing state-of-the-art saliency detection methods heavily rely on CNN-based architectures. Alternatively, we rethink this task from a convolution-free sequence-to-sequence perspective and predict saliency by modeling long-range dependencies, which can not be achieved by convolution. Specifically, we develop a novel unified model based on a pure transformer, namely, Visual Saliency Transformer (VST), for both RGB and RGB-D salient object detection (SOD). It takes image patches as inputs and leverages the transformer to propagate global contexts among image patches. Unlike conventional architectures used in Vision Transformer (ViT), we leverage multi-level token fusion and propose a new token upsampling method under the transformer framework to get high-resolution detection results. We also develop a token-based multi-task decoder to simultaneously perform saliency and boundary detection by introducing task-related tokens and a novel patch-task-attention mechanism. Experimental results show that our model outperforms existing methods on both RGB and RGB-D SOD benchmark datasets. Most importantly, our whole framework not only provides a new perspective for the SOD field but also shows a new paradigm for transformer-based dense prediction models. Code is available at https://github.com/nnizhang/VST.
We propose an approach to generate images of people given a desired appearance and pose. Disentangled representations of pose and appearance are necessary to handle the compound variability in the resulting generated images. Hence, we develop an approach based on intermediate representations of poses and appearance: our pose-guided appearance rendering network firstly encodes the targets' poses using an encoder-decoder neural network. Then the targets' appearances are encoded by learning adaptive appearance filters using a fully convolutional network. Finally, these filters are placed in the encoder-decoder neural networks to complete the rendering. We demonstrate that our model can generate images and videos that are superior to state-of-the-art methods, and can handle pose guided appearance rendering in both image and video generation.
Recently, large-scale transformer-based models have been proven to be effective over various tasks across many domains. Nevertheless, applying them in industrial production requires tedious and heavy works to reduce inference costs. To fill such a gap, we introduce a scalable inference solution: Easy and Efficient Transformer (EET), including a series of transformer inference optimization at the algorithm and implementation levels. First, we design highly optimized kernels for long inputs and large hidden sizes. Second, we propose a flexible CUDA memory manager to reduce the memory footprint when deploying a large model. Compared with the state-of-the-art transformer inference library (Faster Transformer v4.0), EET can achieve an average of 1.40-4.20x speedup on the transformer decoder layer with an A100 GPU
We present a learned, spatially-varying steganography system that allows detecting when and how images have been altered by cropping, splicing or inpainting after publication. The system comprises a learned encoder that imperceptibly hides distinct positional signatures in every local image region before publication, and an accompanying learned decoder that extracts the steganographic signatures to determine, for each local image region, its 2D positional coordinates within the originally-published image. Crop and replacement edits become detectable by the inconsistencies they cause in the hidden positional signatures. Using a prototype system for small $(400 \times 400)$ images, we show experimentally that simple CNN encoder and decoder architectures can be trained jointly to achieve detection that is reliable and robust, without introducing perceptible distortion. This approach could help individuals and image-sharing platforms certify that an image was published by a trusted source, and also know which parts of such an image, if any, have been substantially altered since publication.
The paper proposes a new text recognition network for scene-text images. Many state-of-the-art methods employ the attention mechanism either in the text encoder or decoder for the text alignment. Although the encoder-based attention yields promising results, these schemes inherit noticeable limitations. They perform the feature extraction (FE) and visual attention (VA) sequentially, which bounds the attention mechanism to rely only on the FE final single-scale output. Moreover, the utilization of the attention process is limited by only applying it directly to the single scale feature-maps. To address these issues, we propose a new multi-scale and encoder-based attention network for text recognition that performs the multi-scale FE and VA in parallel. The multi-scale channels also undergo regular fusion with each other to develop the coordinated knowledge together. Quantitative evaluation and robustness analysis on the standard benchmarks demonstrate that the proposed network outperforms the state-of-the-art in most cases.
Despite the significant progress made by deep learning in natural image matting, there has been so far no representative work on deep learning for video matting due to the inherent technical challenges in reasoning temporal domain and lack of large-scale video matting datasets. In this paper, we propose a deep learning-based video matting framework which employs a novel and effective spatio-temporal feature aggregation module (ST-FAM). As optical flow estimation can be very unreliable within matting regions, ST-FAM is designed to effectively align and aggregate information across different spatial scales and temporal frames within the network decoder. To eliminate frame-by-frame trimap annotations, a lightweight interactive trimap propagation network is also introduced. The other contribution consists of a large-scale video matting dataset with groundtruth alpha mattes for quantitative evaluation and real-world high-resolution videos with trimaps for qualitative evaluation. Quantitative and qualitative experimental results show that our framework significantly outperforms conventional video matting and deep image matting methods applied to video in presence of multi-frame temporal information.