Abstract In multiplayer, first‐person shooter games like Counter‐Strike: Global Offensive (CS:GO), coordinated movement is a critical component of high‐level strategic play. However, the complexity of team coordination and the variety of conditions present in popular game maps make it impractical to author hand‐crafted movement policies for every scenario. We show that it is possible to take a data‐driven approach to creating human‐like movement controllers for CS:GO. We curate a team movement dataset comprising 123 hours of professional game play traces, and use this dataset to train a transformer‐based movement model that generates human‐like team movement for all players in a “Retakes” round of the game. Importantly, the movement prediction model is efficient. Performing inference for all players takes less than 0.5 ms per game step (amortized cost) on a single CPU core, making it plausible for use in commercial games today. Human evaluators assess that our model behaves more like humans than both commercially‐available bots and procedural movement controllers scripted by experts (16% to 59% higher by TrueSkill rating of “human‐like”). Using experiments involving in‐game bot vs. bot self‐play, we demonstrate that our model performs simple forms of teamwork, makes fewer common movement mistakes, and yields movement distributions, player lifetimes, and kill locations similar to those observed in professional CS:GO match play.

Many defenses against adversarial attacks (e.g. robust classifiers, randomization, or image purification) use countermeasures put to work only after the attack has been crafted. We adopt a different perspective to introduce A5(Adversarial Augmentation Against Adversarial Attacks), a novel framework including the first certified preemptive defense against adversarial attacks. The main idea is to craft a defensive perturbation to guarantee that any attack (up to a given magnitude) towards the input in hand will fail. To this aim, we leverage existing automatic perturbation analysis tools for neural networks. We study the conditions to apply A5effectively, analyze the importance of the robustness of the to-be-defended classifier, and inspect the appearance of the robustified images. We show effective on-the-fly defensive augmentation with a robustifier network that ignores the ground truth label, and demonstrate the benefits of robustifier and classifier co-training. In our tests, A5consistently beats state of the art certified defenses on MNIST, CIFAR10, FashionMNIST and Tinyimagenet. We also show how to apply A5to create certifiably robust physical objects. Our code at https://github.com/NVlabs/A5 allows experimenting on a wide range of scenarios beyond the man-in-the-middle attack tested here, including the case of physical attacks.

On intelligent vehicles, Deep Neural Networks (DNNs) may run on devices whose computational load varies over time. Within the context of variable network architectures, that can be used to constrain the inference latency for real-time deployment with varying system resources, we introduce LeAF (Legacy Augmentation for Flexible inference), a novel paradigm to augment a pre-trained DNN with trainable, shallow execution paths that can run in place of the legacy ones. While preserving the legacy DNN weights, LeAF allows changing the DNN architecture with minimal overhead to effectively adapt to different system performance targets. LeAF-ResNet-50 has less than 14% storage overhead over the legacy DNN; its accuracy varies from the legacy 76.1% to 70.15% (up to 5% better than Slimmable [1] with a latency that is 37% better than OFA [2] on an A100 GPU with batch size 256). Our analysis shows the importance of considering not only the target device, but also the batch size and the temporal dynamic of the DNN configuration to optimize the performances of variable architecture DNNs, LeAF in particular.

Extracting interesting scenarios from real-world data as well as generating failure cases is important for the development and testing of autonomous systems. We propose efficient mechanisms to both characterize and generate testing scenarios using a state-of-the-art driving simulator. For any scenario, our method generates a set of possible driving paths and identifies all the possible safe driving trajectories that can be taken starting at different times, to compute metrics that quantify the complexity of the scenario. We use our method to characterize real driving data from the Next Generation Simulation (NGSIM) project, as well as adversarial scenarios generated in simulation. We rank the scenarios by defining metrics based on the complexity of avoiding accidents and provide insights into how the AV could have minimized the probability of incurring an accident. We demonstrate a strong correlation between the proposed metrics and human intuition.

Tree Search (TS) is crucial to some of the most influential successes in reinforcement learning. Here, we tackle two major challenges with TS that limit its usability: \textit{distribution shift} and \textit{scalability}. We first discover and analyze a counter-intuitive phenomenon: action selection through TS and a pre-trained value function often leads to lower performance compared to the original pre-trained agent, even when having access to the exact state and reward in future steps. We show this is due to a distribution shift to areas where value estimates are highly inaccurate and analyze this effect using Extreme Value theory. To overcome this problem, we introduce a novel off-policy correction term that accounts for the mismatch between the pre-trained value and its corresponding TS policy by penalizing under-sampled trajectories. We prove that our correction eliminates the above mismatch and bound the probability of sub-optimal action selection. Our correction significantly improves pre-trained Rainbow agents without any further training, often more than doubling their scores on Atari games. Next, we address the scalability issue given by the computational complexity of exhaustive TS that scales exponentially with the tree depth. We introduce Batch-BFS: a GPU breadth-first search that advances all nodes in each depth of the tree simultaneously. Batch-BFS reduces runtime by two orders of magnitude and, beyond inference, enables also training with TS of depths that were not feasible before. We train DQN agents from scratch using TS and show improvement in several Atari games compared to both the original DQN and the more advanced Rainbow. We will share the code upon publication.

We introduce CuLE (CUDA Learning Environment), a CUDA port of the Atari Learning Environment (ALE) which is used for the development of deep reinforcement algorithms. CuLE overcomes many limitations of existing CPU-based emulators and scales naturally to multiple GPUs. It leverages GPU parallelization to run thousands of games simultaneously and it renders frames directly on the GPU, to avoid the bottleneck arising from the limited CPU-GPU communication bandwidth. CuLE generates up to 155M frames per hour on a single GPU, a finding previously achieved only through a cluster of CPUs. Beyond highlighting the differences between CPU and GPU emulators in the context of reinforcement learning, we show how to leverage the high throughput of CuLE by effective batching of the training data, and show accelerated convergence for A2C+V-trace. CuLE is available at https://github.com/NVlabs/cule.

Structural pruning of neural network parameters reduces computational, energy, and memory transfer costs during inference. We propose a novel method that estimates the contribution of a neuron (filter) to the final loss and iteratively removes those with smaller scores. We describe two variations of our method using the first and second-order Taylor expansions to approximate a filter's contribution. Both methods scale consistently across any network layer without requiring per-layer sensitivity analysis and can be applied to any kind of layer, including skip connections. For modern networks trained on ImageNet, we measured experimentally a high (>93%) correlation between the contribution computed by our methods and a reliable estimate of the true importance. Pruning with the proposed methods led to an improvement over state-of-the-art in terms of accuracy, FLOPs, and parameter reduction. On ResNet-101, we achieve a 40% FLOPS reduction by removing 30% of the parameters, with a loss of 0.02% in the top-1 accuracy on ImageNet.

Non-local-means image denoising is based on processing a set of neighbors for a given reference patch. few nearest neighbors (NN) can be used to limit the computational burden of the algorithm. Resorting to a toy problem, we show analytically that sampling neighbors with the NN approach introduces a bias in the denoised patch. We propose a different neighbors' collection criterion to alleviate this issue, which we name statistical NN (SNN). Our approach outperforms the traditional one in case of both white and colored noise: fewer SNNs can be used to generate images of superior quality, at a lower computational cost. A detailed investigation of our toy problem explains the differences between NN and SNN from a grounded point of view. The intuition behind SNN is quite general, and it leads to image quality improvement also in the case of bilateral filtering. The MATLAB code to replicate the results presented in the paper is freely available.

Superpixel generation is a common preprocessing step in vision processing aimed at dividing an image into non-overlapping regions. Simple Linear Iterative Clustering (SLIC) is a commonly used superpixel algorithm that offers a good balance between performance and accuracy. However, the algorithm's high computational and memory bandwidth requirements result in performance and energy efficiency that do not meet the requirements of real-time embedded applications. In this work, we explore the design of an energy-efficient superpixel accelerator for real-time computer vision applications. We propose a novel algorithm, Subsampled SLIC (S-SLIC), that uses pixel subsampling to reduce the memory bandwidth by 1.8×. We integrate S-SLIC into an energy-efficient superpixel accelerator and perform an in-depth design space exploration to optimize the design. We completed a detailed design in a 16nm FinFET technology using commercially-available EDA tools for high-level synthesis to map the design automatically from a C-based representation to a gate-level implementation. The proposed S-SLIC accelerator achieves real-time performance (30 frames per second) with 250× better energy efficiency than an optimized SLIC software implementation running on a mobile GPU.

From self-driving cars to high dynamic range (HDR) imaging, the demand for image-based applications is growing quickly. In mobile systems, these applications place particular strain on performance and energy efficiency. As traditional memory systems are optimized for 1D memory access, they are unable to efficiently exploit the multi-dimensional locality characteristics of image-based applications which often operate on sub-regions of 2D and 3D image data. We have developed a new Patch Memory System (PMEM) tailored to application domains that process 2D and 3D data streams. PMEM supports efficient multidimensional addressing, automatic handling of image boundaries, and efficient caching and prefetching of image data. In addition to an optimized cache, PMEM includes hardware for offloading structured address calculations from processing units. We improve average energy-delay by 26% compared to EVA, a memory system for computer vision applications. Compared to a traditional cache, our results show that PMEM can reduce processor energy by 34% for a selection of CV and IP applications, leading to system performance improvement of up to 32% and energy-delay product improvement of 48-86% on the applications in this study.

We introduce a method for accurate three dimensional head pose estimation using a commodity depth camera. We perform pose estimation by registering a morphable face model to the measured depth data, using a combination of particle swarm optimization (PSO) and the iterative closest point (ICP) algorithm, which minimizes a cost function that includes a 3D registration and a 2D overlap term. The pose is estimated on the fly without requiring an explicit initialization or training phase. Our method handles large pose angles and partial occlusions by dynamically adapting to the reliable visible parts of the face. It is robust and generalizes to different depth sensors without modification. On the Biwi Kinect dataset, we achieve best-in-class performance, with average angular errors of 2.1, 2.1 and 2.4 degrees for yaw, pitch, and roll, respectively, and an average translational error of 5.9 mm, while running at 6 fps on a graphics processing unit.

A new filter to restore radiographic images corrupted by impulsive noise is proposed. It is based on a switching scheme where all the pulses are first detected and then corrected through a median filter. The pulse detector is based on the hypothesis that the major contribution to image noise is given by the photon counting process, with some pixels corrupted by impulsive noise. Such statistics is described by an adequate mixture model. The filter is also able to reliably estimate the sensor gain. Its operation has been verified on both synthetic and real images; the experimental results demonstrate the superiority of the proposed approach in comparison with more traditional methods.

We present a new algorithm, called the soft-tissue filter, that can make both soft and bone tissue clearly visible in digital cephalic radiographies under a wide range of exposures. It uses a mixture model made up of two Gaussian distributions and one inverted lognormal distribution to analyze the image histogram. The image is clustered in three parts: background, soft tissue, and bone using this model. Improvement in the visibility of both structures is achieved through a local transformation based on gamma correction, stretching, and saturation, which is applied using different parameters for bone and soft-tissue pixels. A processing time of 1 s for 5 Mpixel images allows the filter to operate in real time. Although the default value of the filter parameters is adequate for most images, real-time operation allows adjustment to recover under- and overexposed images or to obtain the best quality subjectively. The filter was extensively clinically tested: quantitative and qualitative results are reported here.