Abstract

Self-supervised speech based models have been found to be successful in predicting brain recordings of subjects experiencing naturalistic story listening. Inspired by the recent progress on deep learning models for various speech-processing tasks, existing literature has leveraged pretrained speech Transformer models for brain encoding. However, there is no work on exploring the efficacy of task-specific finetuned Transformer representations for this task. Hence, in this paper, we explore transfer learning from representations finetuned for eight different tasks from Speech processing Universal PERformance Benchmark (SUPERB) for predicting brain responses. Encoding models based on task features are used to predict activity in different regions across the whole brain, and also in language and auditory brain regions. Our experiments on finetuning the Wav2Vec2.0 model for these eight tasks show that the model finetuned on automatic speech recognition (ASR) yields the best encoding performance for the whole brain, language and auditory regions.

Abstract

Coordinate-based implicit neural networks, or neural fields, have emerged as useful representations of shape and appearance in 3D computer vision. Despite advances, however, it remains challenging to build neural fields for categories of objects without datasets like ShapeNet that provide “canonicalized” object instances that are consistently aligned for their 3D position and orientation (pose).We present Canonical Field Network (CaFi-Net), a self-supervised method to canonicalize the 3D pose of instances from an object category represented as neural fields, specifically neural radiance fields (NeRFs). CaFi-Net directly learns from continuous and noisy radiance fields using a Siamese network architecture that is designed to extract equivariant field features for category-level canonicalization. During inference, our method takes pre-trained neural radiance fields of novel object instances at arbitrary 3D pose and estimates a canonical field with consistent 3D pose across the entire category. Extensive experiments on a new dataset of 1300 NeRF models across 13 object categories show that our method matches or exceeds the performance of 3D point cloud-based methods.

Abstract

Multivariate Data Analysis for Motor Failure Detection and Isolation in a Multicopter UAV using Real-Flight Attitude Signals

Abstract

Traffic flow prediction has been regarded as a critical problem in intelligent transportation systems. An accurate prediction can help mitigate congestion and other societal problems while facilitating safer, cost and time-efficient travel. However, this requires the prediction algorithm to consider several complex characteristics of traffic flow data. These complex characteristics are an amalgamation of the spatial, temporal and periodic features exhibited by traffic flow data. To extract and leverage these features for traffic flow prediction, several hybrid deep learning models have been developed recently; however, there are still some challenges to determine the optimal architecture considering both spatial and temporal features. In this work, we perform an extensive comparison of hybrid deep learning models with and without periodicity to understand the prediction accuracy of these popular models. We propose an optimization framework that unifies genetic algorithm (GA) embedded optimization with prediction models in order to derive optimized deep learning architectures for hybrid traffic flow prediction exploring 2D spatial and temporal information. The framework enables the improvement of prediction performance and eliminates the handtuning process. An improved temporal convolutional network (TCN) architecture is derived using the GA driven optimization, which achieves superior traffic flow prediction accuracy compared to all other existing hybrid deep learning models on the freeway and urban traffic data from the PeMS traffic data set. We also evaluate the performance of the derived hybrid deep learning algorithms on the Raspberry PI embedded platform. Index Terms—Traffic flow prediction, Hybrid deep learning models, Temporal convolutional networks

Abstract

Unmanned Aerial Vehicles (UAVs), particularly low-cost UAVs, have become increasingly important due to their wide range of applications and ease of use. However, with the rapid growth of the UAV market, the rising security concerns pose a greater risk. One such primary concern is location spoofing attacks which can compromise UAV's navigation system, making it crucial to analyze various location-based attacks. In this paper, we identify 16 such GPS spoofing scenarios based on environmental conditions, attack type, and spoof signal propagation path. We evaluate these scenarios based on various GPS parameters like Horizontal Dilution Of Precision (HDOP), Vertical Dilution Of Precision (VDOP), GPS satellite count in view, and avg signal-to-noise power density (CN0). We then analyze the variations in GPS parameters for various such attack scenarios. Further, we analyze the impact of distance on average CN0 and the effect of satellite count on effective spoofable distance. We also discuss several critical insights which are empirically observed during our experimental trials. Our experiments revealed that the natural conditions within indoor and outdoor scenarios can vary considerably, and effective spoofable distance can be up to 100 meters when the satellite count is less than 10.

Abstract

Significant progress has been made in speaker dependent Lip-to-Speech synthesis, which aims to generate speech from silent videos of talking faces. Current state-of-the-art approaches primarily employ non-autoregressive sequence-to-sequence architectures to directly predict mel-spectrograms or audio waveforms from lip representations. We hypothesize that the direct mel-prediction hampers training/model efficiency due to the entanglement of speech content with ambient information and speaker characteristics. To this end, we propose RobustL2S, a modularized framework for Lip-to-Speech synthesis. First, a non-autoregressive sequence-to-sequence model maps self-supervised visual features to a representation of disentangled speech content. A vocoder then converts the speech features into raw waveforms. Extensive evaluations confirm the effectiveness of our setup, achieving state-of-the-art performance on the unconstrained Lip2Wav dataset and the constrained GRID and TCD-TIMIT datasets. Speech samples from RobustL2S

Abstract

Efficiently and accurately rendering hair accounting for multiple scattering is a challenging open problem. Path tracing in hair takes long to converge while other techniques are either too approximate while still being computationally expensive or make assumptions about the scene. We present a technique to infer the higher order scattering in hair in constant time within the path tracing framework, while achieving better computational efficiency. Our method makes no assumptions about the scene and provides control over the renderer’s bias & speedup. We achieve this by training a small multilayer perceptron (MLP) to learn the higher-order radiance online, while rendering progresses. We describe how to robustly train this network and thoroughly analyze our resulting renderer’s characteristics. We evaluate our method on various hairstyles and lighting conditions. We also compare our method against a recent learning based & a traditional real-time hair rendering method and demonstrate better quantitative & qualitative results. Our method achieves a significant improvement in speed with respect to path tracing, achieving a run-time reduction of 40% −70% while only introducing a small amount of bias.

Abstract

Text and signs around roads provide crucial information for drivers, vital for safe navigation and situational awareness. Scene text recognition in motion is a challenging problem, while textual cues typically appear for a short time span, and early detection at a distance is necessary. Systems that exploit such information to assist the driver should not only extract and incorporate visual and textual cues from the video stream but also reason over time. To address this issue, we introduce RoadTextVQA, a new dataset for the task of video question answering (VideoQA) in the context of driver assistance. RoadTextVQA consists of 3,222 driving videos collected from multiple countries, annotated with 10,500 questions, all based on text or road signs present in the driving videos. We assess the performance of state-of-the-art video question answering models on our RoadTextVQA dataset, highlighting the significant potential for improvement in this domain and the usefulness of the dataset in advancing research on in-vehicle support systems and text-aware multimodal question answering.

Abstract

A system for generating high-resolution de-warped omni-directional stereo image from captured omni-directional stereo image by correcting optical distortions using projection patterns is provided. The system includes a projection pattern capturing arrangement, a projector or a display, and a de-warping server. The projection pattern capturing arrangement includes one or more omnidirectional cameras to capture projection patterns from the captured omni-directional stereo image from each omni-directional stereo camera. The projector or the display displays the projection patterns. The de-warping server obtain the projection patterns and processes the projection patterns to generate high resolution de-warped omni-directional stereo image by correcting optical distortions in the captured omni-directional stereo image and mapping the captured omni-directional stereo image and the high resolution de-warped omni-directional stereo image.

Abstract

Evaluating the risk level of adversarial images is essential for safely deploying face authentication models in the real world. Popular approaches for physical-world attacks, such as print or replay attacks, suffer from some limitations, like including physical and geometrical artifacts. Recently adversarial attacks have gained attraction, which try to digitally deceive the learning strategy of a recognition system using slight modifications to the captured image. While most previous research assumes that the adversarial image could be digitally fed into the authentication systems, this is not always the case for systems deployed in the real world. This paper demonstrates the vulnerability of face authentication systems to adversarial images in physical world scenarios. We propose AdvGen, an automated Generative Adversarial Network, to simulate print and replay attacks and generate adversarial images that can fool state-of-the-art PADs in a physical domain attack setting. Using this attack strategy, the attack success rate goes up to 82.01%. We test AdvGen extensively on four datasets and ten state-of-the-art PADs. We also demonstrate the effectiveness of our attack by conducting experiments in a realistic, physical environment.