我们为变体视觉任务提供了一个概念上简单，灵活和通用的视觉感知头，例如分类，对象检测，实例分割和姿势估计以及不同的框架，例如单阶段或两个阶段的管道。我们的方法有效地标识了图像中的对象，同时同时生成高质量的边界框或基于轮廓的分割掩码或一组关键点。该方法称为Unihead，将不同的视觉感知任务视为通过变压器编码器体系结构学习的可分配点。给定固定的空间坐标，Unihead将其自适应地分散到了不同的空间点和有关它们的关系的原因。它以多个点的形式直接输出最终预测集，使我们能够在具有相同头部设计的不同框架中执行不同的视觉任务。我们展示了对成像网分类的广泛评估以及可可套件的所有三个曲目，包括对象检测，实例分割和姿势估计。如果没有铃铛和口哨声，Unihead可以通过单个视觉头设计统一这些视觉任务，并与为每个任务开发的专家模型相比，实现可比的性能。我们希望我们的简单和通用的Unihead能够成为可靠的基线，并有助于促进通用的视觉感知研究。代码和型号可在https://github.com/sense-x/unihead上找到。

视觉变压器（VITS）已成为各种视觉任务的流行结构和优于卷积神经网络（CNNS）。然而，这种强大的变形金机带来了巨大的计算负担。而这背后的基本障碍是排气的令牌到令牌比较。为了缓解这一点，我们深入研究Vit的模型属性，观察到VITS表现出稀疏关注，具有高令牌相似性。这直观地向我们介绍了可行的结构不可知的尺寸，令牌编号，以降低计算成本。基于这一探索，我们为香草vits提出了一种通用的自我切片学习方法，即坐下。具体而言，我们首先设计一种新颖的令牌减肥模块（TSM），可以通过动态令牌聚集来提高VIT的推理效率。不同于令牌硬滴，我们的TSM轻轻地集成了冗余令牌变成了更少的信息，可以在不切断图像中的鉴别性令牌关系的情况下动态缩放视觉注意。此外，我们介绍了一种简洁的密集知识蒸馏（DKD）框架，其密集地以柔性自动编码器方式传送无组织的令牌信息。由于教师和学生之间的结构类似，我们的框架可以有效地利用结构知识以获得更好的收敛性。最后，我们进行了广泛的实验来评估我们的坐姿。它展示了我们的方法可以通过1.7倍加速VITS，其精度下降可忽略不计，甚至在3.6倍上加速VITS，同时保持其性能的97％。令人惊讶的是，通过简单地武装LV-VIT与我们的坐线，我们在想象中实现了新的最先进的表现，超过了最近文学中的所有CNN和VITS。

Denoising Diffusion Probabilistic Models (DDPMs) are emerging in text-to-speech (TTS) synthesis because of their strong capability of generating high-fidelity samples. However, their iterative refinement process in high-dimensional data space results in slow inference speed, which restricts their application in real-time systems. Previous works have explored speeding up by minimizing the number of inference steps but at the cost of sample quality. In this work, to improve the inference speed for DDPM-based TTS model while achieving high sample quality, we propose ResGrad, a lightweight diffusion model which learns to refine the output spectrogram of an existing TTS model (e.g., FastSpeech 2) by predicting the residual between the model output and the corresponding ground-truth speech. ResGrad has several advantages: 1) Compare with other acceleration methods for DDPM which need to synthesize speech from scratch, ResGrad reduces the complexity of task by changing the generation target from ground-truth mel-spectrogram to the residual, resulting into a more lightweight model and thus a smaller real-time factor. 2) ResGrad is employed in the inference process of the existing TTS model in a plug-and-play way, without re-training this model. We verify ResGrad on the single-speaker dataset LJSpeech and two more challenging datasets with multiple speakers (LibriTTS) and high sampling rate (VCTK). Experimental results show that in comparison with other speed-up methods of DDPMs: 1) ResGrad achieves better sample quality with the same inference speed measured by real-time factor; 2) with similar speech quality, ResGrad synthesizes speech faster than baseline methods by more than 10 times. Audio samples are available at https://resgrad1.github.io/.

In this work, we investigate improving the generalizability of GAN-generated image detectors by performing data augmentation in the fingerprint domain. Specifically, we first separate the fingerprints and contents of the GAN-generated images using an autoencoder based GAN fingerprint extractor, followed by random perturbations of the fingerprints. Then the original fingerprints are substituted with the perturbed fingerprints and added to the original contents, to produce images that are visually invariant but with distinct fingerprints. The perturbed images can successfully imitate images generated by different GANs to improve the generalization of the detectors, which is demonstrated by the spectra visualization. To our knowledge, we are the first to conduct data augmentation in the fingerprint domain. Our work explores a novel prospect that is distinct from previous works on spatial and frequency domain augmentation. Extensive cross-GAN experiments demonstrate the effectiveness of our method compared to the state-of-the-art methods in detecting fake images generated by unknown GANs.

We study a challenging task, conditional human motion generation, which produces plausible human motion sequences according to various conditional inputs, such as action classes or textual descriptors. Since human motions are highly diverse and have a property of quite different distribution from conditional modalities, such as textual descriptors in natural languages, it is hard to learn a probabilistic mapping from the desired conditional modality to the human motion sequences. Besides, the raw motion data from the motion capture system might be redundant in sequences and contain noises; directly modeling the joint distribution over the raw motion sequences and conditional modalities would need a heavy computational overhead and might result in artifacts introduced by the captured noises. To learn a better representation of the various human motion sequences, we first design a powerful Variational AutoEncoder (VAE) and arrive at a representative and low-dimensional latent code for a human motion sequence. Then, instead of using a diffusion model to establish the connections between the raw motion sequences and the conditional inputs, we perform a diffusion process on the motion latent space. Our proposed Motion Latent-based Diffusion model (MLD) could produce vivid motion sequences conforming to the given conditional inputs and substantially reduce the computational overhead in both the training and inference stages. Extensive experiments on various human motion generation tasks demonstrate that our MLD achieves significant improvements over the state-of-the-art methods among extensive human motion generation tasks, with two orders of magnitude faster than previous diffusion models on raw motion sequences.

Along with the widespread use of face recognition systems, their vulnerability has become highlighted. While existing face anti-spoofing methods can be generalized between attack types, generic solutions are still challenging due to the diversity of spoof characteristics. Recently, the spoof trace disentanglement framework has shown great potential for coping with both seen and unseen spoof scenarios, but the performance is largely restricted by the single-modal input. This paper focuses on this issue and presents a multi-modal disentanglement model which targetedly learns polysemantic spoof traces for more accurate and robust generic attack detection. In particular, based on the adversarial learning mechanism, a two-stream disentangling network is designed to estimate spoof patterns from the RGB and depth inputs, respectively. In this case, it captures complementary spoofing clues inhering in different attacks. Furthermore, a fusion module is exploited, which recalibrates both representations at multiple stages to promote the disentanglement in each individual modality. It then performs cross-modality aggregation to deliver a more comprehensive spoof trace representation for prediction. Extensive evaluations are conducted on multiple benchmarks, demonstrating that learning polysemantic spoof traces favorably contributes to anti-spoofing with more perceptible and interpretable results.

In this work, we present a new computer vision task named video object of interest segmentation (VOIS). Given a video and a target image of interest, our objective is to simultaneously segment and track all objects in the video that are relevant to the target image. This problem combines the traditional video object segmentation task with an additional image indicating the content that users are concerned with. Since no existing dataset is perfectly suitable for this new task, we specifically construct a large-scale dataset called LiveVideos, which contains 2418 pairs of target images and live videos with instance-level annotations. In addition, we propose a transformer-based method for this task. We revisit Swin Transformer and design a dual-path structure to fuse video and image features. Then, a transformer decoder is employed to generate object proposals for segmentation and tracking from the fused features. Extensive experiments on LiveVideos dataset show the superiority of our proposed method.

Error correction in automatic speech recognition (ASR) aims to correct those incorrect words in sentences generated by ASR models. Since recent ASR models usually have low word error rate (WER), to avoid affecting originally correct tokens, error correction models should only modify incorrect words, and therefore detecting incorrect words is important for error correction. Previous works on error correction either implicitly detect error words through target-source attention or CTC (connectionist temporal classification) loss, or explicitly locate specific deletion/substitution/insertion errors. However, implicit error detection does not provide clear signal about which tokens are incorrect and explicit error detection suffers from low detection accuracy. In this paper, we propose SoftCorrect with a soft error detection mechanism to avoid the limitations of both explicit and implicit error detection. Specifically, we first detect whether a token is correct or not through a probability produced by a dedicatedly designed language model, and then design a constrained CTC loss that only duplicates the detected incorrect tokens to let the decoder focus on the correction of error tokens. Compared with implicit error detection with CTC loss, SoftCorrect provides explicit signal about which words are incorrect and thus does not need to duplicate every token but only incorrect tokens; compared with explicit error detection, SoftCorrect does not detect specific deletion/substitution/insertion errors but just leaves it to CTC loss. Experiments on AISHELL-1 and Aidatatang datasets show that SoftCorrect achieves 26.1% and 9.4% CER reduction respectively, outperforming previous works by a large margin, while still enjoying fast speed of parallel generation.

We shed light on a pitfall and an opportunity in physics-informed neural networks (PINNs). We prove that a multilayer perceptron (MLP) only with ReLU (Rectified Linear Unit) or ReLU-like Lipschitz activation functions will always lead to a vanished Hessian. Such a network-imposed constraint contradicts any second- or higher-order partial differential equations (PDEs). Therefore, a ReLU-based MLP cannot form a permissible function space for the approximation of their solutions. Inspired by this pitfall, we prove that a linear PDE up to the $n$-th order can be strictly satisfied by an MLP with $C^n$ activation functions when the weights of its output layer lie on a certain hyperplane, as called the out-layer-hyperplane. An MLP equipped with the out-layer-hyperplane becomes "physics-enforced", no longer requiring a loss function for the PDE itself (but only those for the initial and boundary conditions). Such a hyperplane exists not only for MLPs but for any network architecture tailed by a fully-connected hidden layer. To our knowledge, this should be the first PINN architecture that enforces point-wise correctness of a PDE. We give the closed-form expression of the out-layer-hyperplane for second-order linear PDEs and provide an implementation.

Video super-resolution is one of the most popular tasks on mobile devices, being widely used for an automatic improvement of low-bitrate and low-resolution video streams. While numerous solutions have been proposed for this problem, they are usually quite computationally demanding, demonstrating low FPS rates and power efficiency on mobile devices. In this Mobile AI challenge, we address this problem and propose the participants to design an end-to-end real-time video super-resolution solution for mobile NPUs optimized for low energy consumption. The participants were provided with the REDS training dataset containing video sequences for a 4X video upscaling task. The runtime and power efficiency of all models was evaluated on the powerful MediaTek Dimensity 9000 platform with a dedicated AI processing unit capable of accelerating floating-point and quantized neural networks. All proposed solutions are fully compatible with the above NPU, demonstrating an up to 500 FPS rate and 0.2 [Watt / 30 FPS] power consumption. A detailed description of all models developed in the challenge is provided in this paper.