Zhigang Zhu

Xuan Wang, Zhigang Zhu

Contextual information plays an important role in many computer vision tasks, such as object detection, video action detection, image classification, etc. Recognizing a single object or action out of context could be sometimes very challenging, and context information may help improve the understanding of a scene or an event greatly. Appearance context information, e.g., colors or shapes of the background of an object can improve the recognition accuracy of the object in the scene. Semantic context (e.g. a keyboard on an empty desk vs. a keyboard next to a desktop computer ) will improve accuracy and exclude unrelated events. Context information that are not in the image itself, such as the time or location of an images captured, can also help to decide whether certain event or action should occur. Other types of context (e.g. 3D structure of a building) will also provide additional information to improve the accuracy. In this survey, different context information that has been used in computer vision tasks is reviewed. We categorize context into different types and different levels. We also review available machine learning models and image/video datasets that can employ context information. Furthermore, we compare context based integration and context-free integration in mainly two classes of tasks: image-based and video-based. Finally, this survey is concluded by a set of promising future directions in context learning and utilization.

Xuan Wang, Hao Tang, Zhigang Zhu

Various contextual information has been employed by many approaches for visual detection tasks. However, most of the existing approaches only focus on specific context for specific tasks. In this paper, GMC, a general framework is proposed for multistage context learning and utilization, with various deep network architectures for various visual detection tasks. The GMC framework encompasses three stages: preprocessing, training, and post-processing. In the preprocessing stage, the representation of local context is enhanced by utilizing commonly used labeling standards. During the training stage, semantic context information is fused with visual information, leveraging prior knowledge from the training dataset to capture semantic relationships. In the post-processing stage, general topological relations and semantic masks for stuff are incorporated to enable spatial context reasoning between objects. The proposed framework provides a comprehensive and adaptable solution for context learning and utilization in visual detection scenarios. The framework offers flexibility with user-defined configurations and provide adaptability to diverse network architectures and visual detection tasks, offering an automated and streamlined solution that minimizes user effort and inference time in context learning and reasoning. Experimental results on the visual detection tasks, for storefront object detection, pedestrian detection and COCO object detection, demonstrate that our framework outperforms previous state-of-the-art detectors and transformer architectures. The experiments also demonstrate that three contextual learning components can not only be applied individually and in combination, but can also be applied to various network architectures, and its flexibility and effectiveness in various detection scenarios.

Wenyuan Lia, Xiao-Yun Wang, Zhigang Zhu et al.

In this work, we propose a data-driven image encryption framework that identifies chaotic maps directly from data using the SINDy-PI algorithm. Unlike conventional encryption schemes relying on predefined maps, our method learns the full explicit dynamics -- including cross-terms and higher-order nonlinearities -- from observational data. The validity of this approach is verified on three distinct chaotic systems: the H{é}non map, the three-dimensional logistic map, and the piecewise-linear Lozi map, demonstrating its generality. The encryption key consists solely of initial conditions; the map structure itself becomes data-dependent, introducing an extra layer of security. Moreover, even when the initial conditions are fixed, different training data (e.g., with a tiny noise seed) lead to slightly different maps, which produce completely different ciphertexts (NPCR $\approx 99.6\%$, UACI $\approx 33.5\%$). Numerical experiments on the H{é}non system show near-ideal information entropy ($\approx 8$ bits), negligible inter-pixel correlation, and extreme sensitivity to initial conditions: a perturbation of $10^{-16}$ causes total decryption failure. The scheme resists both differential and statistical attacks, with NPCR and UACI values matching theoretical ideals. Our results establish a new paradigm for chaos-based cryptography beyond fixed maps.

Ling Zhang, Zhigang Zhu

To alleviate the cost of collecting and annotating large-scale point cloud datasets, we propose an unsupervised learning approach to learn features from unlabeled point cloud "3D object" dataset by using part contrasting and object clustering with deep graph neural networks (GNNs). In the contrast learning step, all the samples in the 3D object dataset are cut into two parts and put into a "part" dataset. Then a contrast learning GNN (ContrastNet) is trained to verify whether two randomly sampled parts from the part dataset belong to the same object. In the cluster learning step, the trained ContrastNet is applied to all the samples in the original 3D object dataset to extract features, which are used to group the samples into clusters. Then another GNN for clustering learning (ClusterNet) is trained to predict the cluster ID of all the training samples. The contrasting learning forces the ContrastNet to learn high-level semantic features of objects but probably ignores low-level features, while the ClusterNet improves the quality of learned features by being trained to discover objects that probably belong to the same semantic categories by the use of cluster IDs. We have conducted extensive experiments to evaluate the proposed framework on point cloud classification tasks. The proposed unsupervised learning approach obtained comparable performance to the state-of-the-art unsupervised learning methods that used much more complicated network structures. The code of this work is publicly available via: https://github.com/lingzhang1/ContrastNet.

Jiawei Liu, Enis Berk Çoban, Zarina Schevchenko et al.

Standard training for Multi-modal Large Language Models (MLLMs) involves concatenating non-textual information, like vision or audio, with a text prompt. This approach may not encourage deep integration of modalities, limiting the model's ability to leverage the core language model's reasoning capabilities. This work examined the impact of interleaved instruction tuning in an audio MLLM, where audio tokens are interleaved within the prompt. Using the Listen, Think, and Understand (LTU) model as a testbed, we conduct an experiment using the Synonym and Hypernym Audio Reasoning Dataset (SHARD), our newly created reasoning benchmark for audio-based semantic reasoning focusing on synonym and hypernym recognition. Our findings show that while even zero-shot interleaved prompting improves performance on our reasoning tasks, a small amount of fine-tuning using interleaved training prompts improves the results further, however, at the expense of the MLLM's audio labeling ability.

Jianing Qi, Jiawei Liu, Hao Tang et al.

Vision Language Models (VLMs) excel at identifying and describing objects but often fail at spatial reasoning. We study why VLMs, such as LLaVA, underutilize spatial cues despite having positional encodings and spatially rich vision encoder features. Our analysis reveals a key imbalance: vision token embeddings have much larger norms than text tokens, suppressing LLM's position embedding. To expose this mechanism, we developed three interpretability tools: (1) the Position Sensitivity Index, which quantifies reliance on token order, (2) the Cross Modality Balance, which reveals attention head allocation patterns, and (3) a RoPE Sensitivity probe, which measures dependence on rotary positional embeddings. These tools uncover that vision tokens and system prompts dominate attention. We validated our mechanistic understanding through targeted interventions that predictably restore positional sensitivity. These findings reveal previously unknown failure modes in multimodal attention and demonstrate how interpretability analysis can guide principled improvements.

Jianing Qi, Xi Ye, Hao Tang et al.

Scaling test-time compute brings substantial performance gains for large language models (LLMs). By sampling multiple answers and heuristically aggregate their answers (e.g., either through majority voting or using verifiers to rank the answers), one can achieve consistent performance gains in math domains. In this paper, we propose a new way to leverage such multiple sample set. We train a compact LLM, called Sample Set Aggregator (SSA), that takes a concatenated sequence of multiple samples and output the final answer, optimizing it for the answer accuracy with reinforcement learning. Experiments on five reasoning datasets demonstrate both the efficacy and efficiency of SSA. Notably, SSA improves over naive majority voting by 8% pass@5 on MATH. Furthermore, our 3B SSA surpasses model-based re-ranking with a much larger 72B process reward model. Our analysis also shows promising generalization ability of SSA, across sample set sizes, base model families and scales, and tasks. By separating LLMs to generate answers and LLMs to analyze and aggregate sampled answers, our approach can work with the outputs from premier black box models easily and efficiently.

Jianing Qi, Hao Tang, Zhigang Zhu

We introduce Policy Gradient Guidance (PGG), a simple extension of classifier-free guidance from diffusion models to classical policy gradient methods. PGG augments the policy gradient with an unconditional branch and interpolates conditional and unconditional branches, yielding a test-time control knob that modulates behavior without retraining. We provide a theoretical derivation showing that the additional normalization term vanishes under advantage estimation, leading to a clean guided policy gradient update. Empirically, we evaluate PGG on discrete and continuous control benchmarks. We find that conditioning dropout-central to diffusion guidance-offers gains in simple discrete tasks and low sample regimes, but dropout destabilizes continuous control. Training with modestly larger guidance ($γ>1$) consistently improves stability, sample efficiency, and controllability. Our results show that guidance, previously confined to diffusion policies, can be adapted to standard on-policy methods, opening new directions for controllable online reinforcement learning.

Zihao Zhang, Susan L. Epstein, Casey Breen et al.

This paper introduces ELUA, the Ecological Laboratory for Urban Agriculture, a collaboration among landscape architects, architects and computer scientists who specialize in artificial intelligence, robotics and computer vision. ELUA has two gantry robots, one indoors and the other outside on the rooftop of a 6-story campus building. Each robot can seed, water, weed, and prune in its garden. To support responsive landscape research, ELUA also includes sensor arrays, an AI-powered camera, and an extensive network infrastructure. This project demonstrates a way to integrate artificial intelligence into an evolving urban ecosystem, and encourages landscape architects to develop an adaptive design framework where design becomes a long-term engagement with the environment.

Xingye Li, Ling Zhang, Zhigang Zhu

Manually annotating complex scene point cloud datasets is both costly and error-prone. To reduce the reliance on labeled data, a new model called SnapshotNet is proposed as a self-supervised feature learning approach, which directly works on the unlabeled point cloud data of a complex 3D scene. The SnapshotNet pipeline includes three stages. In the snapshot capturing stage, snapshots, which are defined as local collections of points, are sampled from the point cloud scene. A snapshot could be a view of a local 3D scan directly captured from the real scene, or a virtual view of such from a large 3D point cloud dataset. Snapshots could also be sampled at different sampling rates or fields of view (FOVs), thus multi-FOV snapshots, to capture scale information from the scene. In the feature learning stage, a new pre-text task called multi-FOV contrasting is proposed to recognize whether two snapshots are from the same object or not, within the same FOV or across different FOVs. Snapshots go through two self-supervised learning steps: the contrastive learning step with both part and scale contrasting, followed by a snapshot clustering step to extract higher level semantic features. Then a weakly-supervised segmentation stage is implemented by first training a standard SVM classifier on the learned features with a small fraction of labeled snapshots. The trained SVM is used to predict labels for input snapshots and predicted labels are converted into point-wise label assignments for semantic segmentation of the entire scene using a voting procedure. The experiments are conducted on the Semantic3D dataset and the results have shown that the proposed method is capable of learning effective features from snapshots of complex scene data without any labels. Moreover, the proposed method has shown advantages when comparing to the SOA method on weakly-supervised point cloud semantic segmentation.

Jie Wei, Zhigang Zhu, Erik Blasch et al.

During natural disasters, aircraft and satellites are used to survey the impacted regions. Usually human experts are needed to manually label the degrees of the building damage so that proper humanitarian assistance and disaster response (HADR) can be achieved, which is labor-intensive and time-consuming. Expecting human labeling of major disasters over a wide area gravely slows down the HADR efforts. It is thus of crucial interest to take advantage of the cutting-edge Artificial Intelligence and Machine Learning techniques to speed up the natural infrastructure damage assessment process to achieve effective HADR. Accordingly, the paper demonstrates a systematic effort to achieve efficient building damage classification. First, two novel generative adversarial nets (GANs) are designed to augment data used to train the deep-learning-based classifier. Second, a contrastive learning based method using novel data structures is developed to achieve great performance. Third, by using information fusion, the classifier is effectively trained with very few training data samples for transfer learning. All the classifiers are small enough to be loaded in a smart phone or simple laptop for first responders. Based on the available overhead imagery dataset, results demonstrate data and computational efficiency with 10% of the collected data combined with a GAN reducing the time of computation from roughly half a day to about 1 hour with roughly similar classification performances.

Greg Olmschenk, Hao Tang, Zhigang Zhu

Gatherings of thousands to millions of people frequently occur for an enormous variety of events, and automated counting of these high-density crowds is useful for safety, management, and measuring significance of an event. In this work, we show that the regularly accepted labeling scheme of crowd density maps for training deep neural networks is less effective than our alternative inverse k-nearest neighbor (i$k$NN) maps, even when used directly in existing state-of-the-art network structures. We also provide a new network architecture MUD-i$k$NN, which uses multi-scale upsampling via transposed convolutions to take full advantage of the provided i$k$NN labeling. This upsampling combined with the i$k$NN maps further improves crowd counting accuracy. Our new network architecture performs favorably in comparison with the state-of-the-art. However, our labeling and upsampling techniques are generally applicable to existing crowd counting architectures.

Greg Olmschenk, Zhigang Zhu, Hao Tang

In this work, we generalize semi-supervised generative adversarial networks (GANs) from classification problems to regression problems. In the last few years, the importance of improving the training of neural networks using semi-supervised training has been demonstrated for classification problems. We present a novel loss function, called feature contrasting, resulting in a discriminator which can distinguish between fake and real data based on feature statistics. This method avoids potential biases and limitations of alternative approaches. The generalization of semi-supervised GANs to the regime of regression problems of opens their use to countless applications as well as providing an avenue for a deeper understanding of how GANs function. We first demonstrate the capabilities of semi-supervised regression GANs on a toy dataset which allows for a detailed understanding of how they operate in various circumstances. This toy dataset is used to provide a theoretical basis of the semi-supervised regression GAN. We then apply the semi-supervised regression GANs to a number of real-world computer vision applications: age estimation, driving steering angle prediction, and crowd counting from single images. We perform extensive tests of what accuracy can be achieved with significantly reduced annotated data. Through the combination of the theoretical example and real-world scenarios, we demonstrate how semi-supervised GANs can be generalized to regression problems.

Wei Li, Farnaz Abitahi, Zhigang Zhu

Action Unit (AU) detection becomes essential for facial analysis. Many proposed approaches face challenging problems in dealing with the alignments of different face regions, in the effective fusion of temporal information, and in training a model for multiple AU labels. To better address these problems, we propose a deep learning framework for AU detection with region of interest (ROI) adaptation, integrated multi-label learning, and optimal LSTM-based temporal fusing. First, ROI cropping nets (ROI Nets) are designed to make sure specifically interested regions of faces are learned independently; each sub-region has a local convolutional neural network (CNN) - an ROI Net, whose convolutional filters will only be trained for the corresponding region. Second, multi-label learning is employed to integrate the outputs of those individual ROI cropping nets, which learns the inter-relationships of various AUs and acquires global features across sub-regions for AU detection. Finally, the optimal selection of multiple LSTM layers to form the best LSTM Net is carried out to best fuse temporal features, in order to make the AU prediction the most accurate. The proposed approach is evaluated on two popular AU detection datasets, BP4D and DISFA, outperforming the state of the art significantly, with an average improvement of around 13% on BP4D and 25% on DISFA, respectively.

Wei Li, Farnaz Abtahi, Zhigang Zhu et al.

In this paper, we propose a deep learning based approach for facial action unit detection by enhancing and cropping the regions of interest. The approach is implemented by adding two novel nets (layers): the enhancing layers and the cropping layers, to a pretrained CNN model. For the enhancing layers, we designed an attention map based on facial landmark features and applied it to a pretrained neural network to conduct enhanced learning (The E-Net). For the cropping layers, we crop facial regions around the detected landmarks and design convolutional layers to learn deeper features for each facial region (C-Net). We then fuse the E-Net and the C-Net to obtain our Enhancing and Cropping (EAC) Net, which can learn both feature enhancing and region cropping functions. Our approach shows significant improvement in performance compared to the state-of-the-art methods applied to BP4D and DISFA AU datasets.

Wei Li, Zhigang Zhu

We propose a deep learning-based feature fusion approach for facial computing including face recognition as well as gender, race and age detection. Instead of training a single classifier on face images to classify them based on the features of the person whose face appears in the image, we first train four different classifiers for classifying face images based on race, age, gender and identification (ID). Multi-task features are then extracted from the trained models and cross-task-feature training is conducted which shows the value of fusing multimodal features extracted from multi-tasks. We have found that features trained for one task can be used for other related tasks. More interestingly, the features trained for a task with more classes (e.g. ID) and then used in another task with fewer classes (e.g. race) outperforms the features trained for the other task itself. The final feature fusion is performed by combining the four types of features extracted from the images by the four classifiers. The feature fusion approach improves the classifications accuracy by a 7.2%, 20.1%, 22.2%, 21.8% margin, respectively, for ID, age, race and gender recognition, over the results of single classifiers trained only on their individual features. The proposed method can be applied to applications in which different types of data or features can be extracted.

Wei Li, Christina Tsangouri, Farnaz Abtahi et al.

In this paper, we propose a recursive framework to recognize facial expressions from images in real scenes. Unlike traditional approaches that typically focus on developing and refining algorithms for improving recognition performance on an existing dataset, we integrate three important components in a recursive manner: facial dataset generation, facial expression recognition model building, and interactive interfaces for testing and new data collection. To start with, we first create a candid-images-for-facial-expression (CIFE) dataset. We then apply a convolutional neural network (CNN) to CIFE and build a CNN model for web image expression classification. In order to increase the expression recognition accuracy, we also fine-tune the CNN model and thus obtain a better CNN facial expression recognition model. Based on the fine-tuned CNN model, we design a facial expression game engine and collect a new and more balanced dataset, GaMo. The images of this dataset are collected from the different expressions our game users make when playing the game. Finally, we evaluate the GaMo and CIFE datasets and show that our recursive framework can help build a better facial expression model for dealing with real scene facial expression tasks.

Wei Li, Farnaz Abtahi, Christina Tsangouri et al.

In order to create an "in-the-wild" dataset of facial emotions with large number of balanced samples, this paper proposes a game-based data collection framework. The framework mainly include three components: a game engine, a game interface, and a data collection and evaluation module. We use a deep learning approach to build an emotion classifier as the game engine. Then a emotion web game to allow gamers to enjoy the games, while the data collection module obtains automatically-labelled emotion images. Using our game, we have collected more than 15,000 images within a month of the test run and built an emotion dataset "GaMo". To evaluate the dataset, we compared the performance of two deep learning models trained on both GaMo and CIFE. The results of our experiments show that because of being large and balanced, GaMo can be used to build a more robust emotion detector than the emotion detector trained on CIFE, which was used in the game engine to collect the face images.