With the rise of deep learning, many companies have developed domain-specific architectures (DSAs) optimized for AI workloads, with Ascend being a representative. To fully realize the operator performance on Ascend, effective analysis and optimization is urgently needed. Compared to GPU, Ascend requires users to manage operations manually, leading to complex performance issues that require precise analysis. However, existing roofline models face challenges of visualization complexity and inaccurate performance assessment. To address these needs, we introduce a component-based roofline model that abstracts components to capture operator performance, thereby effectively identifying bottleneck components. Furthermore, through practical operator optimization case studies, we illustrate a comprehensive process of optimization based on roofline analysis, summarizing common performance issues and optimization strategies. Finally, extensive end-to-end optimization experiments demonstrate significant model speed improvements, ranging from 1.07× to 2.15×, along with valuable insights from practice.

Offline model-based optimization (MBO) aims to identify a design that maximizes a black-box function using only a fixed, pre-collected dataset of designs and their corresponding scores. This problem has garnered significant attention from both scientific and industrial domains. A common approach in offline MBO is to train a regression-based surrogate model by minimizing mean squared error (MSE) and then find the best design within this surrogate model by different optimizers (e.g., gradient ascent). However, a critical challenge is the risk of out-of-distribution errors, i.e., the surrogate model may typically overestimate the scores and mislead the optimizers into suboptimal regions. Prior works have attempted to address this issue in various ways, such as using regularization techniques and ensemble learning to enhance the robustness of the model, but it still remains. In this paper, we argue that regression models trained with MSE are not well-aligned with the primary goal of offline MBO, which is to \textit{select} promising designs rather than to predict their scores precisely. Notably, if a surrogate model can maintain the order of candidate designs based on their relative score relationships, it can produce the best designs even without precise predictions. To validate it, we conduct experiments to compare the relationship between the quality of the final designs and MSE, finding that the correlation is really very weak. In contrast, a metric that measures order-maintaining quality shows a significantly stronger correlation. Based on this observation, we propose learning a ranking-based model that leverages learning to rank techniques to prioritize promising designs based on their relative scores. We show that the generalization error on ranking loss can be well bounded. Empirical results across diverse tasks demonstrate the superior performance of our proposed ranking-based method than twenty existing methods. Our implementation is available at \url{https://github.com/lamda-bbo/Offline-RaM}.

The pursuit of universal black-box optimization (BBO) algorithms is a longstanding goal. However, unlike domains such as language or vision, where scaling structured data has driven generalization, progress in offline BBO remains hindered by the lack of unified representations for heterogeneous numerical spaces. Thus, existing offline BBO approaches are constrained to single-task and fixed-dimensional settings, failing to achieve cross-domain universal optimization. Recent advances in language models (LMs) offer a promising path forward: their embeddings capture latent relationships in a unifying way, enabling universal optimization across different data types possible. In this paper, we discuss multiple potential approaches, including an end-to-end learning framework in the form of next-token prediction, as well as prioritizing the learning of latent spaces with strong representational capabilities. To validate the effectiveness of these methods, we collect offline BBO tasks and data from open-source academic works for training. Experiments demonstrate the universality and effectiveness of our proposed methods. Our findings suggest that unifying language model priors and learning string embedding space can overcome traditional barriers in universal BBO, paving the way for general-purpose BBO algorithms. The code is provided at https://github.com/lamda-bbo/universal-offline-bbo.

High-speed object tracking holds significant relevance across robotic domains, such as drones and autonomous driving. Compared to conventional cameras, event cameras are equipped with the ability to capture object motion information at exceptionally high temporal resolution with relatively low power consumption and remain immune from motion-blurring effects. Regrettably, many existing methods adopt a framebased approach by stacking events into Event Frame, which overlooks the sparsity and high temporal resolution of events. This approach is also reliant on the huge pre-training backbone and reaches a performance plateau but demands unrealistically large networks and high power consumption, rendering it impractical for real-time applications in battery-constrained robotic scenarios. In this paper, we propose an efficient and effective single-modality tracker using Point Cloud representation named E2B (Event to Box). By directly handling the raw output of event cameras without dataformat transformation, E2B leverages events' coordinate guidance to accurately map Event Cloud features to 2D bounding boxes. Moreover, E2B incorporates the pyramid structure into the multi-stage feature extraction architecture to effectively track objects across diverse scales. In the experiments, E2B performs outstandingly on two large-scale and one synthetic event-based tracking datasets, covering both indoor and outdoor environments, as well as rigid and non-rigid objects.

In crowd navigation, the local goal plays a crucial role in trajectory initialization, optimization, and evaluation. Recognizing that when the global goal is distant, the robot's primary objective is avoiding collisions, making it less critical to pass through the exact local goal point, this work introduces the concept of goal lines, which extend the traditional local goal from a single point to multiple candidate lines. Coupled with a topological map construction strategy that groups obstacles to be as convex as possible, a goal-adaptive navigation framework is proposed to efficiently plan multiple candidate trajectories. Simulations and experiments demonstrate that the proposed GA-TEB framework effectively prevents deadlock situations, where the robot becomes frozen due to a lack of feasible trajectories in crowded environments. Additionally, the framework greatly increases planning frequency in scenarios with numerous non-convex obstacles, enhancing both robustness and safety.

The performance of an algorithm often critically depends on its hyperparameter configuration. Dynamic algorithm configuration (DAC) is a recent trend in automated machine learning, which can dynamically adjust the algorithm’s configuration during the execution process and relieve users from tedious trial-and-error tuning tasks. Recently, multi-agent reinforcement learning (MARL) approaches have improved the configuration of multiple heterogeneous hyperparameters, making various parameter configurations for complex algorithms possible. However, many complex algorithms have inherent inter-dependencies among multiple parameters (e.g., determining the operator type first and then the operator's parameter), which are, however, not considered in previous approaches, thus leading to sub-optimal results. In this paper, we propose the sequential multi-agent DAC (Seq-MADAC) framework to address this issue by considering the inherent inter-dependencies of multiple parameters. Specifically, we propose a sequential advantage decomposition network, which can leverage action-order information through sequential advantage decomposition. Experiments from synthetic functions to the configuration of multi-objective optimization algorithms demonstrate Seq-MADAC's superior performance over state-of-the-art MARL methods and show strong generalization across problem classes. Seq-MADAC establishes a new paradigm for the widespread dependency-aware automated algorithm configuration. Our code is available at https://github.com/lamda-bbo/seq-madac.

Recovering a sequence of latent sharp frames from a motion-blurred image is a challenging task. The bio-inspired event camera, which produces an event stream with high temporal resolution, has been exploited to promote the recovery performance. However, recovering sharp sequences with arbitrary temporal scales has been ignored for a long time. Existing works can only recover a fixed number of latent frames from a blurry image once they are trained. In this work, we propose an event-assisted blurry image unfolding framework that can work across arbitrary temporal scales. A bi-directional recurrent network is employed to encode events corresponding to each latent frame, which gathers information over all events in the exposure time. Features of both the blurry image and events are fused together and fed to a bi-directional latent sequence decoder (BiLSD) to produce a sequence of latent sharp frames. Extensive experiments show that the proposed method not only performs favorably against state-of-the-art methods in recovering a fixed number of frames from a blurry image but can be well generalized to arbitrary-temporal-scale blurry image unfolding.

Recently audio-driven talking face video generation has attracted considerable attention. However, very few researches address the issue of emotional editing of these talking face videos with continuously controllable expressions, which is a strong demand in the industry. The challenge is that speech-related expressions and emotion-related expressions are often highly coupled. Meanwhile, traditional image-to-image translation methods cannot work well in our application due to the coupling of expressions with other attributes such as poses, i.e., translating the expression of the character in each frame may simultaneously change the head pose due to the bias of the training data distribution. In this paper, we propose a high-quality facial expression editing method for talking face videos, allowing the user to control the target emotion in the edited video continuously. We present a new perspective for this task as a special case of motion information editing, where we use a 3DMM to capture major facial movements and an associated texture map modeled by a StyleGAN to capture appearance details. Both representations (3DMM and texture map) contain emotional information and can be continuously modified by neural networks and easily smoothed by averaging in coefficient/latent spaces, making our method simple yet effective. We also introduce a mouth shape preservation loss to control the trade-off between lip synchronization and the degree of exaggeration of the edited expression. Extensive experiments and a user study show that our method achieves state-of-the-art performance across various evaluation criteria.

The goal of blurry image deblurring and unfolding task is to recover a single sharp frame or a sequence from a blurry one. Recently, its performance is greatly improved with introduction of a bio-inspired visual sensor, event camera. Most existing event-assisted deblurring methods focus on the design of powerful network architectures and effective training strategy, while ignoring the role of blur modeling in removing various blur in dynamic scenes. In this work, we propose to implicitly model blur in an image by computing blurriness representation with an event-assisted blurriness encoder. The learning of blurriness representation is formulated as a ranking problem based on specially synthesized pairs. Blurriness-aware image unfolding is achieved by integrating blur relevant information contained in the representation into a base unfolding network. The integration is mainly realized by the proposed blurriness-guided modulation and multi-scale aggregation modules. Experiments on GOPRO and HQF datasets show favorable performance of the proposed method against state-of-the-art approaches. More results on real-world data validate its effectiveness in recovering a sequence of latent sharp frames from a blurry image.

Event-based cameras are bio-inspired sensors that capture brightness change of every pixel in an asynchronous manner. Compared with frame-based sensors, event cameras have microsecond-level latency and high dynamic range, hence showing great potential for object detection under high-speed motion and poor illumination conditions. Due to sparsity and asynchronism nature with event streams, most of existing approaches resort to hand-crafted methods to convert event data into 2D grid representation. However, they are sub-optimal in aggregating information from event stream for object detection. In this work, we propose to learn an event representation optimized for event-based object detection. Specifically, event streams are divided into grids in the x-y-t coordinates for both positive and negative polarity, producing a set of pillars as 3D tensor representation. To fully exploit information with event streams to detect objects, a dual-memory aggregation network (DMANet) is proposed to leverage both long and short memory along event streams to aggregate effective information for object detection. Long memory is encoded in the hidden state of adaptive convLSTMs while short memory is modeled by computing spatial-temporal correlation between event pillars at neighboring time intervals. Extensive experiments on the recently released event-based automotive detection dataset demonstrate the effectiveness of the proposed method.

Deploying deep learning models to new environments is very challenging. Domain adaptation (DA) is a promising paradigm to solve the problem by collecting and adapting to unlabeled data in new environments. Though research efforts have led to steady performance improvement over the past decade, DA algorithms are still hard to deploy, as training on unlabeled new data makes tuning difficult and not feasible for inference-only devices. To make DA practical, in this paper we study a new problem named Test-time Training-Free Domain Adaptation (TTDA), where trained models must adapt to a single input (mimicking the test-time scenario) without training. By exploiting spatial activation that was previously overlooked and simply averaged out, we propose a simple method based on Feature Statistics Transformation (FST) on-the-fly for each test example. The proposed algorithm is tested in the TTDA setting on two standard DA benchmarks. Surprisingly, it surpasses or performs on par with state-of-the-art DA methods, even though they require additional training. We envision that this training-free paradigm has the potential to bring DA to embedded devices and would be of interest to audience of community.

Generating audio-driven photo-realistic talking face has received intensive attention due to its ability to bring more new human-computer interaction experiences. However, previous works struggled to balance high definition, lip synchronization, and low customization costs, which would degrade the user experience. In this paper, a novel audio-driven talking face generation method was proposed, which subtly converts the problem of improving video definition into the problem of face reenactment to produce both lip-synchronized and high- definition face video. The framework is decoupled, meaning that the same trained model can be used on arbitrary characters and audio without further customizing training for specific people, thus significantly reducing costs. Experiment results show that our proposed method achieves the high video definition, and comparable lip synchronization performance with the existing state-of-the-art methods.

Spiking Neural Networks (SNNs) are well known as a promising energy-efficient alternative to conventional artificial neural networks. Subject to the preconceived impression that SNNs are sparse firing, the analysis and optimization of inherent redundancy in SNNs have been largely overlooked, thus the potential advantages of spike-based neuromorphic computing in accuracy and energy efficiency are interfered. In this work, we pose and focus on three key questions regarding the inherent redundancy in SNNs. We argue that the redundancy is induced by the spatio-temporal invariance of SNNs, which enhances the efficiency of parameter utilization but also invites lots of noise spikes. Further, we analyze the effect of spatio-temporal invariance on the spatio-temporal dynamics and spike firing of SNNs. Then, motivated by these analyses, we propose an Advance Spatial Attention (ASA) module to harness SNNs’ redundancy, which can adaptively optimize their membrane potential distribution by a pair of individual spatial attention sub-modules. In this way, noise spike features are accurately regulated. Experimental results demonstrate that the proposed method can significantly drop the spike firing with better performance than state-of-the-art SNN baselines. Our code is available in https://github.com/BICLab/ASA-SNN.

In the past years, attention-based Transformers have swept across the field of computer vision, starting a new stage of backbones in semantic segmentation. Nevertheless, semantic segmentation under poor light conditions remains an open problem. Moreover, most papers about semantic segmentation work on images produced by commodity frame-based cameras with a limited framerate, hindering their deployment to auto-driving systems that require instant perception and response at milliseconds. An event camera is a new sensor that generates event data at microseconds and can work in poor light conditions with a high dynamic range. It looks promising to leverage event cameras to enable perception where commodity cameras are incompetent, but algorithms for event data are far from mature. Pioneering researchers stack event data as frames so that event-based segmentation is converted to frame-based segmentation, but characteristics of event data are not explored. Noticing that event data naturally highlight moving objects, we propose a posterior attention module that adjusts the standard attention by the prior knowledge provided by event data. The posterior attention module can be readily plugged into many segmentation backbones. Plugging the posterior attention module into a recently proposed SegFormer network, we get EvSegFormer (the event-based version of SegFormer) with state-of-the-art performance in two datasets (MVSEC and DDD-17) collected for event-based segmentation. Code is available at https://github.com/zexiJia/EvSegFormer to facilitate research on event-based vision.

Recording fast motion in a high FPS (frame-per-second) requires expensive high-speed cameras. As an alternative, interpolating low-FPS videos from commodity cameras has attracted significant attention. If only low-FPS videos are available, motion assumptions (linear or quadratic) are necessary to infer intermediate frames, which fail to model complex motions. Event camera, a new camera with pixels producing events of brightness change at the temporal resolution of μs (10–6second), is a game-changing device to enable video interpolation at the presence of arbitrarily complex motion. Since event camera is a novel sensor, its potential has not been fulfilled due to the lack of processing algorithms. The pioneering work Time Lens introduced event cameras to video interpolation by designing optical devices to collect a large amount of paired training data of high-speed frames and events, which is too costly to scale. To fully unlock the potential of event cameras, this paper proposes a novel TimeReplayer algorithm to interpolate videos captured by commodity cameras with events. It is trained in an unsupervised cycleconsistent style, canceling the necessity of high-speed training data and bringing the additional ability of video extrapolation. Its state-of-the-art results and demo videos in supplementary reveal the promising future of event-based vision.

Audio-driven talking face, driving talking face by audio, has received considerable attention in multi-modal learning due to its widespread use in virtual reality. However, long-time recording of target high-quality video is needed by most existing audio-driven talking face studies, which significantly increases customization costs. This paper proposes a novel data-efficient audio-driven talking face generation method, which uses just a short target video to produce both lip-synchronized and high-definition face video driven by arbitrary audio in the wild. Current methods suffer from many problems, such as low definition, asynchronization of lip movement and voice, and intense demands for videos for training. In this work, the original target character’s face images are decomposed into 3D face model parameters including expression, geometry, illumination, etc. Then, low-definition pseudo video generated by an adapted target face video bridges the powerful pre-trained audio-driven model to our audio-to-expression transformation network and help to transfer the ability of audio-identity disentanglement. The expression is replaced via an audio and then combined with other face parameters to render a synthetic face. Finally, a neural rendering network translates the synthetic face into talking face without loss of definition. Experimental results show that the proposed method has the best performance in high-definition image quality, and comparable performance in lip synchronization compared with the existing state-of-the-art methods.