Zi Wang

Dustin Tran, Jeremiah Liu, Michael W. Dusenberry et al. · oxford

A recent trend in artificial intelligence is the use of pretrained models for language and vision tasks, which have achieved extraordinary performance but also puzzling failures. Probing these models' abilities in diverse ways is therefore critical to the field. In this paper, we explore the reliability of models, where we define a reliable model as one that not only achieves strong predictive performance but also performs well consistently over many decision-making tasks involving uncertainty (e.g., selective prediction, open set recognition), robust generalization (e.g., accuracy and proper scoring rules such as log-likelihood on in- and out-of-distribution datasets), and adaptation (e.g., active learning, few-shot uncertainty). We devise 10 types of tasks over 40 datasets in order to evaluate different aspects of reliability on both vision and language domains. To improve reliability, we developed ViT-Plex and T5-Plex, pretrained large model extensions for vision and language modalities, respectively. Plex greatly improves the state-of-the-art across reliability tasks, and simplifies the traditional protocol as it improves the out-of-the-box performance and does not require designing scores or tuning the model for each task. We demonstrate scaling effects over model sizes up to 1B parameters and pretraining dataset sizes up to 4B examples. We also demonstrate Plex's capabilities on challenging tasks including zero-shot open set recognition, active learning, and uncertainty in conversational language understanding.

Yichao Yan, Zanwei Zhou, Zi Wang et al.

Conversation is an essential component of virtual avatar activities in the metaverse. With the development of natural language processing, textual and vocal conversation generation has achieved a significant breakthrough. However, face-to-face conversations account for the vast majority of daily conversations, while most existing methods focused on single-person talking head generation. In this work, we take a step further and consider generating realistic face-to-face conversation videos. Conversation generation is more challenging than single-person talking head generation, since it not only requires generating photo-realistic individual talking heads but also demands the listener to respond to the speaker. In this paper, we propose a novel unified framework based on neural radiance field (NeRF) to address this task. Specifically, we model both the speaker and listener with a NeRF framework, with different conditions to control individual expressions. The speaker is driven by the audio signal, while the response of the listener depends on both visual and acoustic information. In this way, face-to-face conversation videos are generated between human avatars, with all the interlocutors modeled within the same network. Moreover, to facilitate future research on this task, we collect a new human conversation dataset containing 34 clips of videos. Quantitative and qualitative experiments evaluate our method in different aspects, e.g., image quality, pose sequence trend, and naturalness of the rendering videos. Experimental results demonstrate that the avatars in the resulting videos are able to perform a realistic conversation, and maintain individual styles. All the code, data, and models will be made publicly available.

Shaoheng Fang, Zi Wang, Yiqi Zhong et al.

Vision-centric joint perception and prediction (PnP) has become an emerging trend in autonomous driving research. It predicts the future states of the traffic participants in the surrounding environment from raw RGB images. However, it is still a critical challenge to synchronize features obtained at multiple camera views and timestamps due to inevitable geometric distortions and further exploit those spatial-temporal features. To address this issue, we propose a temporal bird's-eye-view pyramid transformer (TBP-Former) for vision-centric PnP, which includes two novel designs. First, a pose-synchronized BEV encoder is proposed to map raw image inputs with any camera pose at any time to a shared and synchronized BEV space for better spatial-temporal synchronization. Second, a spatial-temporal pyramid transformer is introduced to comprehensively extract multi-scale BEV features and predict future BEV states with the support of spatial-temporal priors. Extensive experiments on nuScenes dataset show that our proposed framework overall outperforms all state-of-the-art vision-based prediction methods.

Yuhe Ding, Jian Liang, Bo Jiang et al.

CLIP-based domain generalization aims to improve model generalization to unseen domains by leveraging the powerful zero-shot classification capabilities of CLIP and multiple source datasets. Existing methods typically train a single model across multiple source domains to capture domain-shared information. However, this paradigm inherently suffers from two types of conflicts: 1) sample conflicts, arising from noisy samples and extreme domain shifts among sources; and 2) optimization conflicts, stemming from competition and trade-offs during multi-source training. Both hinder the generalization and lead to suboptimal solutions. Recent studies have shown that model merging can effectively mitigate the competition of multi-objective optimization and improve generalization performance. Inspired by these findings, we propose Harmonizing and Merging (HAM), a novel source model merging framework for CLIP-based domain generalization. During the training process of the source models, HAM enriches the source samples without conflicting samples, and harmonizes the update directions of all models. Then, a redundancy-aware historical model merging method is introduced to effectively integrate knowledge across all source models. HAM comprehensively consolidates source domain information while enabling mutual enhancement among source models, ultimately yielding a final model with optimal generalization capabilities. Extensive experiments on five widely used benchmark datasets demonstrate the effectiveness of our approach, achieving state-of-the-art performance.

Aihua Zheng, Ziling He, Zi Wang et al.

Many existing multi-modality studies are based on the assumption of modality integrity. However, the problem of missing arbitrary modalities is very common in real life, and this problem is less studied, but actually important in the task of multi-modality person re-identification (Re-ID). To this end, we design a novel dynamic enhancement network (DENet), which allows missing arbitrary modalities while maintaining the representation ability of multiple modalities, for partial multi-modality person Re-ID. To be specific, the multi-modal representation of the RGB, near-infrared (NIR) and thermal-infrared (TIR) images is learned by three branches, in which the information of missing modalities is recovered by the feature transformation module. Since the missing state might be changeable, we design a dynamic enhancement module, which dynamically enhances modality features according to the missing state in an adaptive manner, to improve the multi-modality representation. Extensive experiments on multi-modality person Re-ID dataset RGBNT201 and vehicle Re-ID dataset RGBNT100 comparing to the state-of-the-art methods verify the effectiveness of our method in complex and changeable environments.

Aihua Zheng, Zhiqi Ma, Zi Wang et al.

Multi-spectral vehicle re-identification aims to address the challenge of identifying vehicles in complex lighting conditions by incorporating complementary visible and infrared information. However, in harsh environments, the discriminative cues in RGB and NIR modalities are often lost due to strong flares from vehicle lamps or sunlight, and existing multi-modal fusion methods are limited in their ability to recover these important cues. To address this problem, we propose a Flare-Aware Cross-modal Enhancement Network that adaptively restores flare-corrupted RGB and NIR features with guidance from the flare-immunized thermal infrared spectrum. First, to reduce the influence of locally degraded appearance due to intense flare, we propose a Mutual Flare Mask Prediction module to jointly obtain flare-corrupted masks in RGB and NIR modalities in a self-supervised manner. Second, to use the flare-immunized TI information to enhance the masked RGB and NIR, we propose a Flare-Aware Cross-modal Enhancement module that adaptively guides feature extraction of masked RGB and NIR spectra with prior flare-immunized knowledge from the TI spectrum. Third, to extract common informative semantic information from RGB and NIR, we propose an Inter-modality Consistency loss that enforces semantic consistency between the two modalities. Finally, to evaluate the proposed FACENet in handling intense flare, we introduce a new multi-spectral vehicle re-ID dataset, called WMVEID863, with additional challenges such as motion blur, significant background changes, and particularly intense flare degradation. Comprehensive experiments on both the newly collected dataset and public benchmark multi-spectral vehicle re-ID datasets demonstrate the superior performance of the proposed FACENet compared to state-of-the-art methods, especially in handling strong flares. The code and dataset will be released at this link.

Chengcheng Li, Zi Wang, Hairong Qi

Building on top of the success of generative adversarial networks (GANs), conditional GANs attempt to better direct the data generation process by conditioning with certain additional information. Inspired by the most recent AC-GAN, in this paper we propose a fast-converging conditional GAN (FC-GAN). In addition to the real/fake classifier used in vanilla GANs, our discriminator has an advanced auxiliary classifier which distinguishes each real class from an extra `fake' class. The `fake' class avoids mixing generated data with real data, which can potentially confuse the classification of real data as AC-GAN does, and makes the advanced auxiliary classifier behave as another real/fake classifier. As a result, FC-GAN can accelerate the process of differentiation of all classes, thus boost the convergence speed. Experimental results on image synthesis demonstrate our model is competitive in the quality of images generated while achieving a faster convergence rate.

Zi Wang, Dali Wang, Chengcheng Li et al.

Cell movement in the early phase of C. elegans development is regulated by a highly complex process in which a set of rules and connections are formulated at distinct scales. Previous efforts have shown that agent-based, multi-scale modeling systems can integrate physical and biological rules and provide new avenues to study developmental systems. However, the application of these systems to model cell movement is still challenging and requires a comprehensive understanding of regulation networks at the right scales. Recent developments in deep learning and reinforcement learning provide an unprecedented opportunity to explore cell movement using 3D time-lapse images. We present a deep reinforcement learning approach within an ABM system to characterize cell movement in C. elegans embryogenesis. Our modeling system captures the complexity of cell movement patterns in the embryo and overcomes the local optimization problem encountered by traditional rule-based, ABM that uses greedy algorithms. We tested our model with two real developmental processes: the anterior movement of the Cpaaa cell via intercalation and the rearrangement of the left-right asymmetry. In the first case, model results showed that Cpaaa's intercalation is an active directional cell movement caused by the continuous effects from a longer distance, as opposed to a passive movement caused by neighbor cell movements. This is because the learning-based simulation found that a passive movement model could not lead Cpaaa to the predefined destination. In the second case, a leader-follower mechanism well explained the collective cell movement pattern. These results showed that our approach to introduce deep reinforcement learning into ABM can test regulatory mechanisms by exploring cell migration paths in a reverse engineering perspective. This model opens new doors to explore large datasets generated by live imaging.

Zi Wang, Chengtao Li, Stefanie Jegelka et al.

Optimization of high-dimensional black-box functions is an extremely challenging problem. While Bayesian optimization has emerged as a popular approach for optimizing black-box functions, its applicability has been limited to low-dimensional problems due to its computational and statistical challenges arising from high-dimensional settings. In this paper, we propose to tackle these challenges by (1) assuming a latent additive structure in the function and inferring it properly for more efficient and effective BO, and (2) performing multiple evaluations in parallel to reduce the number of iterations required by the method. Our novel approach learns the latent structure with Gibbs sampling and constructs batched queries using determinantal point processes. Experimental validations on both synthetic and real-world functions demonstrate that the proposed method outperforms the existing state-of-the-art approaches.