Mean field approximation methodology has laid the foundation of modern Continuous Random Field (CRF) based solutions for the refinement of semantic segmentation. In this paper, we propose to relax the hard constraint of mean field approximation - minimizing the energy term of each node from probabilistic graphical model, by a global optimization with the proposed dilated sparse convolution module (DSConv). In addition, adaptive global average-pooling and adaptive global max-pooling are implemented as replacements of fully connected layers. In order to integrate DSConv, we design an end-to-end, time-efficient DilatedCRF pipeline. The unary energy term is derived either from pre-softmax and post-softmax features, or the predicted affordance map using a conventional classifier, making it easier to implement DilatedCRF for varieties of classifiers. We also present superior experimental results of proposed approach on the suction dataset comparing to other CRF-based approaches.
Motion, as the most distinct phenomenon in a video to involve the changes over time, has been unique and critical to the development of video representation learning. In this paper, we ask the question: how important is the motion particularly for self-supervised video representation learning. To this end, we compose a duet of exploiting the motion for data augmentation and feature learning in the regime of contrastive learning. Specifically, we present a Motion-focused Contrastive Learning (MCL) method that regards such duet as the foundation. On one hand, MCL capitalizes on optical flow of each frame in a video to temporally and spatially sample the tubelets (i.e., sequences of associated frame patches across time) as data augmentations. On the other hand, MCL further aligns gradient maps of the convolutional layers to optical flow maps from spatial, temporal and spatio-temporal perspectives, in order to ground motion information in feature learning. Extensive experiments conducted on R(2+1)D backbone demonstrate the effectiveness of our MCL. On UCF101, the linear classifier trained on the representations learnt by MCL achieves 81.91% top-1 accuracy, outperforming ImageNet supervised pre-training by 6.78%. On Kinetics-400, MCL achieves 66.62% top-1 accuracy under the linear protocol. Code is available at https://github.com/YihengZhang-CV/MCL-Motion-Focused-Contrastive-Learning.
From the original theoretically well-defined spectral graph convolution to the subsequent spatial bassed message-passing model, spatial locality (in vertex domain) acts as a fundamental principle of most graph neural networks (GNNs). In the spectral graph convolution, the filter is approximated by polynomials, where a $k$-order polynomial covers $k$-hop neighbors. In the message-passing, various definitions of neighbors used in aggregations are actually an extensive exploration of the spatial locality information. For learning node representations, the topological distance seems necessary since it characterizes the basic relations between nodes. However, for learning representations of the entire graphs, is it still necessary to hold? In this work, we show that such a principle is not necessary, it hinders most existing GNNs from efficiently encoding graph structures. By removing it, as well as the limitation of polynomial filters, the resulting new architecture significantly boosts performance on learning graph representations. We also study the effects of graph spectrum on signals and interpret various existing improvements as different spectrum smoothing techniques. It serves as a spatial understanding that quantitatively measures the effects of the spectrum to input signals in comparison to the well-known spectral understanding as high/low-pass filters. More importantly, it sheds the light on developing powerful graph representation models.
Building extraction from fine-resolution remote sensing images plays a vital role in numerous geospatial applications, such as urban planning, population statistic, economic assessment and disaster management. With the advancement of deep learning technology, deep convolutional neural networks (DCNNs) have dominated the automatic building extraction task for many years. However, the local property of DCNNs limits the extraction of global information, weakening the ability of the network for recognizing the building instance. Recently, the Transformer comprises a hot topic in the computer vision domain and achieves state-of-the-art performance in fundamental vision tasks, such as image classification, semantic segmentation and object detection. Inspired by this, in this paper, we propose a novel transformer-based network for extracting buildings from fine-resolution remote sensing images, namely BuildFormer. In Comparision with the ResNet, the proposed method achieves an improvement of 2% in mIoU on the WHU building dataset.
Simultaneous reconstruction of geometry and reflectance properties in uncontrolled environments remains a challenging problem. In this paper, we propose an efficient method to reconstruct the scene's 3D geometry and reflectance from multi-view photography using conventional hand-held cameras. Our method automatically builds a virtual scene in a differentiable rendering system that roughly matches the real world's scene parameters, optimized by minimizing photometric objectives alternatingly and stochastically. With the optimal scene parameters evaluated, photo-realistic novel views for various viewing angles and distances can then be generated by our approach. We present the results of captured scenes with complex geometry and various reflection types. Our method also shows superior performance compared to state-of-the-art alternatives in novel view synthesis visually and quantitatively.
In the digital world, signals are discrete and finite. The Fourier representation of discrete and finite signals is FT convolution of the finite sampling function and the continuous signal. Conventionally, finite sampling is treated as a segment of infinite sampling. Though this approach perfectly solves the difference between finite and infinite sampling, it has caused much trouble for signal processing. Mathematically, there is a kind of sampling between finite and infinite sampling, and we name this kind of sampling as half-infinite sampling. Theoretically, finite sampling can also be treated as a segment of half-infinite sampling. Because we can derive the Fourier representation of discrete and finite signals from half-infinite sampling, the FTs of several half-infinite samplings are studied. The results show that the FT of half-infinite sampling is more concise than that of infinite sampling. A numerical experiment verified the theoretical derivations successfully, during which step sampling was proposed. Besides, several interesting equations were built.
Semantic segmentation of fine-resolution urban scene images plays a vital role in extensive practical applications, such as land cover mapping, urban change detection, environmental protection and economic assessment. Driven by rapid developments in deep learning technologies, the convolutional neural network (CNN) has dominated the semantic segmentation task for many years. Convolutional neural networks adopt hierarchical feature representation, demonstrating strong local information extraction. However, the local property of the convolution layer limits the network from capturing global context that is crucial for precise segmentation. Recently, Transformer comprise a hot topic in the computer vision domain. Transformer demonstrates the great capability of global information modelling, boosting many vision tasks, such as image classification, object detection and especially semantic segmentation. In this paper, we propose an efficient hybrid Transformer (EHT) for real-time urban scene segmentation. The EHT adopts a hybrid structure with and CNN-based encoder and a transformer-based decoder, learning global-local context with lower computation. Extensive experiments demonstrate that our EHT has faster inference speed with competitive accuracy compared with state-of-the-art lightweight models. Specifically, the proposed EHT achieves a 66.9% mIoU on the UAVid test set and outperforms other benchmark networks significantly. The code will be available soon.
One-shot voice cloning aims to transform speaker voice and speaking style in speech synthesized from a text-to-speech (TTS) system, where only a shot recording from the target reference speech can be used. Out-of-domain transfer is still a challenging task, and one important aspect that impacts the accuracy and similarity of synthetic speech is the conditional representations carrying speaker or style cues extracted from the limited references. In this paper, we present a novel one-shot voice cloning algorithm called Unet-TTS that has good generalization ability for unseen speakers and styles. Based on a skip-connected U-net structure, the new model can efficiently discover speaker-level and utterance-level spectral feature details from the reference audio, enabling accurate inference of complex acoustic characteristics as well as imitation of speaking styles into the synthetic speech. According to both subjective and objective evaluations of similarity, the new model outperforms both speaker embedding and unsupervised style modeling (GST) approaches on an unseen emotional corpus.
Consider a home or office where multiple devices are running voice assistants (e.g., TVs, lights, ovens, refrigerators, etc.). A human user turns to a particular device and gives a voice command, such as ``Alexa, can you ...''. This paper focuses on the problem of detecting which device the user was facing, and therefore, enabling only that device to respond to the command. Our core intuition emerges from the fact that human voice exhibits a directional radiation pattern, and the orientation of this pattern should influence the signal received at each device. Unfortunately, indoor multipath, unknown user location, and unknown voice signals pose as critical hurdles. Through a new algorithm that estimates the line-of-sight (LoS) power from a given signal, and combined with beamforming and triangulation, we design a functional solution called CoDIR. Results from $500+$ configurations, across $5$ rooms and $9$ different users, are encouraging. While improvements are necessary, we believe this is an important step forward in a challenging but urgent problem space.