Search for: All records

Neighborhood attention reduces the cost of self attention by restricting each token’s attention span to its nearest neighbors. This restriction, parameterized by a window size and dilation factor, draws a spectrum of possible attention patterns between linear projection and self attention. Neighborhood attention, and more generally sliding window attention patterns, have long been bounded by infrastructure, particularly in higher-rank spaces (2-D and 3-D), calling for the development of custom kernels, which have been limited in either functionality, or performance, if not both. In this work, we aim to massively improve upon existing infrastructure by providing two new methods for implementing neighborhood attention. We first show that neighborhood attention can be represented as a batched GEMM problem, similar to standard attention, and implement it for 1-D and 2-D neighborhood attention. These kernels on average provide 895% and 272% improvement in full precision runtime compared to existing naive CUDA kernels for 1-D and 2-D neighborhood attention respectively. We find that aside from being heavily bound by memory bandwidth, certain inherent inefficiencies exist in all unfused implementations of neighborhood attention, which in most cases undo their theoretical efficiency gain. Motivated by the progress made into fused dot-product attention kernels, we developed fused neighborhood attention; an adaptation of fused dot-product attention kernels that allow fine-grained control over attention across different spatial axes. Known for reducing the quadratic time complexity of self attention to a linear complexity, neighborhood attention can now enjoy a reduced and constant memory footprint, and record-breaking half precision runtime. We observe that our fused implementation successfully circumvents some of the unavoidable inefficiencies in unfused implementations. While our unfused GEMM-based kernels only improve half precision performance compared to naive kernels by an average of 548% and 193% in 1-D and 2-D problems respectively, our fused kernels improve naive kernels by an average of 1759% and 958% in 1-D and 2-D problems respectively. These improvements translate into up to 104% improvement in inference and 39% improvement in training existing models based on neighborhood attention, and additionally extend its applicability to image and video perception, as well as other modalities. 

This paper presents a novel material spectroscopy approach to facial presentation–attack–defense (PAD). Best-in-class PAD methods typically detect artifacts in the 3D space. This paper proposes similar features can be achieved in a monocular, single-frame approach by using controlled light. A mathematical model is produced to show how live faces and their spoof counterparts have unique reflectance patterns due to geometry and albedo. A rigorous dataset is collected to evaluate this proposal: 30 diverse adults and their spoofs (paper-mask, display-replay, spandex-mask and COVID mask) under varied pose, position, and lighting for 80,000 unique frames. A panel of 13 texture classifiers are then benchmarked to verify the hypothesis. The experimental results are excellent. The material spectroscopy process enables a conventional MobileNetV3 network to achieve 0.8% average-classification-error rate, outperforming the selected state-of-the-art algorithms. This demonstrates the proposed imaging methodology generates extremely robust features. 

Face-swap-attacks (FSAs) are a new threat to face recognition systems. FSAs are essentially imperceptible replay-attacks using an injection device and generative networks. By placing the device between the camera and computer device, attackers can present any face as desired. This is particularly potent as it also maintains liveliness features, as it is a sophisticated alternation of a real person, and as it can go undetected by traditional anti-spoofing methods. To address FSAs, this research proposes a noise-verification framework. Even the best generative networks today leave alteration traces in the photo-response noise profile; these are detected by doing a comparison of challenge images against the camera enrollment. This research also introduces compression and sub-zone analysis for efficiency. Benchmarking with open-source tampering-detection algorithms shows the proposed compressed-PRNU verification robustly verifies facial-image authenticity while being significantly faster. This demonstrates a novel efficiency for mitigating face-swap-attacks, including denial-of-service attacks. 

This paper describes the derivation, analysis and implementation of a new data association method that provides a tight bound on the risk of incorrect association for LiDAR feature-based localization. Data association (DA) is the process of assigning currently-sensed features with ones that were previously observed. Most DA methods use a nearest-neighbor criterion based on the normalized innovation squared (NIS). They require complex algorithms to evaluate the risk of incorrect association because sensor state prediction, prior observations, and current measurements are uncertain. In contrast, in this work, we derive a new DA criterion using projections of the extended Kalman filter's innovation vector. The paper shows that innovation projections (IP) are signed quantities that not only capture the impact of an incorrect association in terms of its magnitude, but also of its direction. The IP-based DA criterion also leverages the fact that incorrect associations are known and well-defined fault modes. Thus, as compared to NIS, IPs provide a much tighter bound on the predicted risk of incorrect association. We analyze and evaluate the new IP method using simulated and experimental data for autonomous inertial-aided LiDAR localization in a structured lab environment.