The Mixture of Experts (MoE) paradigm provides a powerful way to decompose inscrutable dense layers into smaller, modular computations often more amenable to human interpretation, debugging, and editability. A major problem however lies in the computational cost of scaling the number of experts to achieve sufficiently fine-grained specialization. In this paper, we propose the Multilinear Mixutre of Experts (MMoE) layer to address this, focusing on vision models. MMoE layers perform an implicit computation on prohibitively large weight tensors entirely in factorized form. Consequently, MMoEs both (1) avoid the issues incurred through the discrete expert routing in the popular 'sparse' MoE models, yet (2) do not incur the restrictively high inference-time costs of 'soft' MoE alternatives. We present both qualitative and quantitative evidence (through visualization and counterfactual interventions respectively) that scaling MMoE layers when fine-tuning foundation models for vision tasks leads to more specialized experts at the class-level whilst remaining competitive with the performance of parameter-matched linear layer counterparts. Finally, we show that learned expert specialism further facilitates manual correction of demographic bias in CelebA attribute classification. Our MMoE model code is available at https://github.com/james-oldfield/MMoE.
We introduce an approach for 3D head avatar generation and editing with multi-modal conditioning based on a 3D Generative Adversarial Network (GAN) and a Latent Diffusion Model (LDM). 3D GANs can generate high-quality head avatars given a single or no condition. However, it is challenging to generate samples that adhere to multiple conditions of different modalities. On the other hand, LDMs excel at learning complex conditional distributions. To this end, we propose to exploit the conditioning capabilities of LDMs to enable multi-modal control over the latent space of a pre-trained 3D GAN. Our method can generate and edit 3D head avatars given a mixture of control signals such as RGB input, segmentation masks, and global attributes. This provides better control over the generation and editing of synthetic avatars both globally and locally. Experiments show that our proposed approach outperforms a solely GAN-based approach both qualitatively and quantitatively on generation and editing tasks. To the best of our knowledge, our approach is the first to introduce multi-modal conditioning to 3D avatar generation and editing. \\href{avatarmmc-sig24.github.io}{Project Page}
Frame quality deterioration is one of the main challenges in the field of video understanding. To compensate for the information loss caused by deteriorated frames, recent approaches exploit transformer-based integration modules to obtain spatio-temporal information. However, these integration modules are heavy and complex. Furthermore, each integration module is specifically tailored for its target task, making it difficult to generalise to multiple tasks. In this paper, we present a neat and unified framework, called Spatio-Temporal Prompting Network (STPN). It can efficiently extract robust and accurate video features by dynamically adjusting the input features in the backbone network. Specifically, STPN predicts several video prompts containing spatio-temporal information of neighbour frames. Then, these video prompts are prepended to the patch embeddings of the current frame as the updated input for video feature extraction. Moreover, STPN is easy to generalise to various video tasks because it does not contain task-specific modules. Without bells and whistles, STPN achieves state-of-the-art performance on three widely-used datasets for different video understanding tasks, i.e., ImageNetVID for video object detection, YouTubeVIS for video instance segmentation, and GOT-10k for visual object tracking. Code is available at https://github.com/guanxiongsun/vfe.pytorch.
Face Recognition (FR) systems can suffer from physical (i.e., print photo) and digital (i.e., DeepFake) attacks. However, previous related work rarely considers both situations at the same time. This implies the deployment of multiple models and thus more computational burden. The main reasons for this lack of an integrated model are caused by two factors: (1) The lack of a dataset including both physical and digital attacks with ID consistency which means the same ID covers the real face and all attack types; (2) Given the large intra-class variance between these two attacks, it is difficult to learn a compact feature space to detect both attacks simultaneously. To address these issues, we collect a Unified physical-digital Attack dataset, called UniAttackData. The dataset consists of $1,800$ participations of 2 and 12 physical and digital attacks, respectively, resulting in a total of 29,706 videos. Then, we propose a Unified Attack Detection framework based on Vision-Language Models (VLMs), namely UniAttackDetection, which includes three main modules: the Teacher-Student Prompts (TSP) module, focused on acquiring unified and specific knowledge respectively; the Unified Knowledge Mining (UKM) module, designed to capture a comprehensive feature space; and the Sample-Level Prompt Interaction (SLPI) module, aimed at grasping sample-level semantics. These three modules seamlessly form a robust unified attack detection framework. Extensive experiments on UniAttackData and three other datasets demonstrate the superiority of our approach for unified face attack detection.
We address the challenging problem of Long-Tailed Semi-Supervised Learning (LTSSL) where labeled data exhibit imbalanced class distribution and unlabeled data follow an unknown distribution. Unlike in balanced SSL, the generated pseudo-labels are skewed towards head classes, intensifying the training bias. Such a phenomenon is even amplified as more unlabeled data will be mislabeled as head classes when the class distribution of labeled and unlabeled datasets are mismatched. To solve this problem, we propose a novel method named ComPlementary Experts (CPE). Specifically, we train multiple experts to model various class distributions, each of them yielding high-quality pseudo-labels within one form of class distribution. Besides, we introduce Classwise Batch Normalization for CPE to avoid performance degradation caused by feature distribution mismatch between head and non-head classes. CPE achieves state-of-the-art performances on CIFAR-10-LT, CIFAR-100-LT, and STL-10-LT dataset benchmarks. For instance, on CIFAR-10-LT, CPE improves test accuracy by over >2.22% compared to baselines. Code is available at https://github.com/machengcheng2016/CPE-LTSSL.
Sign Languages (SL) serve as the predominant mode of communication for the Deaf and Hard of Hearing communities. The advent of deep learning has aided numerous methods in SL recognition and translation, achieving remarkable results. However, Sign Language Production (SLP) poses a challenge for the computer vision community as the motions generated must be realistic and have precise semantic meanings. Most SLP methods rely on 2D data, thus impeding their ability to attain a necessary level of realism. In this work, we propose a diffusion-based SLP model trained on a curated large-scale dataset of 4D signing avatars and their corresponding text transcripts. The proposed method can generate dynamic sequences of 3D avatars from an unconstrained domain of discourse using a diffusion process formed on a novel and anatomically informed graph neural network defined on the SMPL-X body skeleton. Through a series of quantitative and qualitative experiments, we show that the proposed method considerably outperforms previous methods of SLP. We believe that this work presents an important and necessary step towards realistic neural sign avatars, bridging the communication gap between Deaf and hearing communities. The code, method and generated data will be made publicly available.
Detecting objects in low-light scenarios presents a persistent challenge, as detectors trained on well-lit data exhibit significant performance degradation on low-light data due to the low visibility. Previous methods mitigate this issue by investigating image enhancement or object detection techniques using low-light image datasets. However, the progress is impeded by the inherent difficulties associated with collecting and annotating low-light images. To address this challenge, we propose to boost low-light object detection with zero-shot day-night domain adaptation, which aims to generalize a detector from well-lit scenarios to low-light ones without requiring real low-light data. We first design a reflectance representation learning module to learn Retinex-based illumination invariance in images with a carefully designed illumination invariance reinforcement strategy. Next, an interchange-redecomposition-coherence procedure is introduced to improve over the vanilla Retinex image decomposition process by performing two sequential image decompositions and introducing a redecomposition cohering loss. Extensive experiments on ExDark, DARK FACE and CODaN datasets show strong low-light generalizability of our method.
Heterogeneous face recognition (HFR) involves the intricate task of matching face images across the visual domains of visible (VIS) and near-infrared (NIR). While much of the existing literature on HFR identifies the domain gap as a primary challenge and directs efforts towards bridging it at either the input or feature level, our work deviates from this trend. We observe that large neural networks, unlike their smaller counterparts, when pre-trained on large scale homogeneous VIS data, demonstrate exceptional zero-shot performance in HFR, suggesting that the domain gap might be less pronounced than previously believed. By approaching the HFR problem as one of low-data fine-tuning, we introduce a straightforward framework: comprehensive pre-training, succeeded by a regularized fine-tuning strategy, that matches or surpasses the current state-of-the-art on four publicly available benchmarks. Corresponding codes can be found at https://github.com/michaeltrs/RethinkNIRVIS.
Stable diffusion, a generative model used in text-to-image synthesis, frequently encounters resolution-induced composition problems when generating images of varying sizes. This issue primarily stems from the model being trained on pairs of single-scale images and their corresponding text descriptions. Moreover, direct training on images of unlimited sizes is unfeasible, as it would require an immense number of text-image pairs and entail substantial computational expenses. To overcome these challenges, we propose a two-stage pipeline named Any-Size-Diffusion (ASD), designed to efficiently generate well-composed images of any size, while minimizing the need for high-memory GPU resources. Specifically, the initial stage, dubbed Any Ratio Adaptability Diffusion (ARAD), leverages a selected set of images with a restricted range of ratios to optimize the text-conditional diffusion model, thereby improving its ability to adjust composition to accommodate diverse image sizes. To support the creation of images at any desired size, we further introduce a technique called Fast Seamless Tiled Diffusion (FSTD) at the subsequent stage. This method allows for the rapid enlargement of the ASD output to any high-resolution size, avoiding seaming artifacts or memory overloads. Experimental results on the LAION-COCO and MM-CelebA-HQ benchmarks demonstrate that ASD can produce well-structured images of arbitrary sizes, cutting down the inference time by 2x compared to the traditional tiled algorithm.