Yunzhong Hou

Yunzhong Hou, Stephen Gould, Liang Zheng

Multiview camera setups have proven useful in many computer vision applications for reducing ambiguities, mitigating occlusions, and increasing field-of-view coverage. However, the high computational cost associated with multiple views poses a significant challenge for end devices with limited computational resources. To address this issue, we propose a view selection approach that analyzes the target object or scenario from given views and selects the next best view for processing. Our approach features a reinforcement learning based camera selection module, MVSelect, that not only selects views but also facilitates joint training with the task network. Experimental results on multiview classification and detection tasks show that our approach achieves promising performance while using only 2 or 3 out of N available views, significantly reducing computational costs. Furthermore, analysis on the selected views reveals that certain cameras can be shut off with minimal performance impact, shedding light on future camera layout optimization for multiview systems. Code is available at https://github.com/hou-yz/MVSelect.

Yunzhong Hou, Stephen Gould, Liang Zheng

Semi-supervised semantic segmentation needs rich and robust supervision on unlabeled data. Consistency learning enforces the same pixel to have similar features in different augmented views, which is a robust signal but neglects relationships with other pixels. In comparison, contrastive learning considers rich pairwise relationships, but it can be a conundrum to assign binary positive-negative supervision signals for pixel pairs. In this paper, we take the best of both worlds and propose multi-view correlation consistency (MVCC) learning: it considers rich pairwise relationships in self-correlation matrices and matches them across views to provide robust supervision. Together with this correlation consistency loss, we propose a view-coherent data augmentation strategy that guarantees pixel-pixel correspondence between different views. In a series of semi-supervised settings on two datasets, we report competitive accuracy compared with the state-of-the-art methods. Notably, on Cityscapes, we achieve 76.8% mIoU with 1/8 labeled data, just 0.6% shy from the fully supervised oracle.

Pengrui Lu, Shiqi Zhang, Yunzhong Hou et al.

Recent coding agents can generate complete codebases from simple prompts, yet existing evaluations focus on issue-level bug fixing and lag behind end-to-end development. We introduce ProjDevBench, an end-to-end benchmark that provides project requirements to coding agents and evaluates the resulting repositories. Combining Online Judge (OJ) testing with LLM-assisted code review, the benchmark evaluates agents on (1) system architecture design, (2) functional correctness, and (3) iterative solution refinement. We curate 20 programming problems across 8 categories, covering both concept-oriented tasks and real-world application scenarios, and evaluate six coding agents built on different LLM backends. Our evaluation reports an overall acceptance rate of 27.38%: agents handle basic functionality and data structures but struggle with complex system design, time complexity optimization, and resource management. Our benchmark is available at https://github.com/zsworld6/projdevbench.

Ziheng Zhang, Yunzhong Hou, Naijing Liu et al.

This paper introduces WARDEN, an early language model system capable of transcribing and translating Wardaman, an endangered Australian indigenous language into English. The significant challenge we face is the lack of large-scale training data: in fact, we only have 6 hours of annotated audio. Therefore, while it is common practice to train a single model for transcription and translation using large datasets (like English to French), this practice is no longer viable in the Wardaman to English context. To tackle the low-resource challenge, we design WARDEN to have separate transcription and translation models: WARDEN first turns a Wardaman audio input into phonemic transcription, and then the transcription into English translation. Further, we propose two useful techniques to enhance performance. For transcription, we initialize the Wardaman token from Sundanese, a language that shares similar phonemes with Wardaman, to accelerate fine-tuning of the transcription model. For translation, we compile a Wardaman-English dictionary from expert annotations, and provide this domain-specific knowledge to a large language model (LLM) to reason and decide the final output. We empirically demonstrate that this two-stage design works better than data-hungry unified approaches in extremely low data settings. Using a mere 6 hours of annotated data, WARDEN outperforms larger open-source and proprietary models and establishes a strong baseline. Data and code are available.

Xingjian Leng, Jaskirat Singh, Yunzhong Hou et al.

In this paper we tackle a fundamental question: "Can we train latent diffusion models together with the variational auto-encoder (VAE) tokenizer in an end-to-end manner?" Traditional deep-learning wisdom dictates that end-to-end training is often preferable when possible. However, for latent diffusion transformers, it is observed that end-to-end training both VAE and diffusion-model using standard diffusion-loss is ineffective, even causing a degradation in final performance. We show that while diffusion loss is ineffective, end-to-end training can be unlocked through the representation-alignment (REPA) loss -- allowing both VAE and diffusion model to be jointly tuned during the training process. Despite its simplicity, the proposed training recipe (REPA-E) shows remarkable performance; speeding up diffusion model training by over 17x and 45x over REPA and vanilla training recipes, respectively. Interestingly, we observe that end-to-end tuning with REPA-E also improves the VAE itself; leading to improved latent space structure and downstream generation performance. In terms of final performance, our approach sets a new state-of-the-art; achieving FID of 1.12 and 1.69 with and without classifier-free guidance on ImageNet 256 x 256. Code is available at https://end2end-diffusion.github.io.

Yunzhong Hou, Liang Zheng, Stephen Gould

Color and structure are the two pillars that combine to give an image its meaning. Interested in critical structures for neural network recognition, we isolate the influence of colors by limiting the color space to just a few bits, and find structures that enable network recognition under such constraints. To this end, we propose a color quantization network, ColorCNN, which learns to structure an image in limited color spaces by minimizing the classification loss. Building upon the architecture and insights of ColorCNN, we introduce ColorCNN+, which supports multiple color space size configurations, and addresses the previous issues of poor recognition accuracy and undesirable visual fidelity under large color spaces. Via a novel imitation learning approach, ColorCNN+ learns to cluster colors like traditional color quantization methods. This reduces overfitting and helps both visual fidelity and recognition accuracy under large color spaces. Experiments verify that ColorCNN+ achieves very competitive results under most circumstances, preserving both key structures for network recognition and visual fidelity with accurate colors. We further discuss differences between key structures and accurate colors, and their specific contributions to network recognition. For potential applications, we show that ColorCNNs can be used as image compression methods for network recognition.

Yuwei Yang, Zeyu Zhang, Yunzhong Hou et al.

Being able to effectively read scientific plots, or chart understanding, is a central part toward building effective agents for science. However, existing multimodal large language models (MLLMs), especially open-source ones, are still falling behind with a typical success rate of 30%-50% on challenging benchmarks. Previous studies on fine-tuning MLLMs with synthetic charts are often restricted by their inadequate similarity to the real charts, which could compromise model training and performance on complex real-world charts. In this study, we show that modularizing chart generation and diversifying visual details improves chart understanding capabilities. In particular, we design a five-step data synthesis pipeline, where we separate data and function creation for single plot generation, condition the generation of later subplots on earlier ones for multi-subplot figures, visually diversify the generated figures, filter out low quality data, and finally generate the question-answer (QA) pairs with GPT-4o. This approach allows us to streamline the generation of fine-tuning datasets and introduce the effective chart dataset (ECD), which contains 10k+ chart images and 300k+ QA pairs, covering 25 topics and featuring 250+ chart type combinations with high visual complexity. We show that ECD consistently improves the performance of various MLLMs on a range of real-world and synthetic test sets. Code, data and models are available at: https://github.com/yuweiyang-anu/ECD.

Yejie Guo, Yunzhong Hou, Wufei Ma et al.

Spatial reasoning, the ability to ground language in 3D understanding, remains a persistent challenge for Vision-Language Models (VLMs). We identify two fundamental bottlenecks: inadequate 3D understanding capabilities stemming from 2D-centric pre-training, and reasoning failures induced by redundant 3D information. To address these, we first construct a Minimal Sufficient Set (MSS) of information before answering a given question: a compact selection of 3D perception results from \textit{expert models}. We introduce MSSR (Minimal Sufficient Spatial Reasoner), a dual-agent framework that implements this principle. A Perception Agent programmatically queries 3D scenes using a versatile perception toolbox to extract sufficient information, including a novel SOG (Situated Orientation Grounding) module that robustly extracts language-grounded directions. A Reasoning Agent then iteratively refines this information to pursue minimality, pruning redundant details and requesting missing ones in a closed loop until the MSS is curated. Extensive experiments demonstrate that our method, by explicitly pursuing both sufficiency and minimality, significantly improves accuracy and achieves state-of-the-art performance across two challenging benchmarks. Furthermore, our framework produces interpretable reasoning paths, offering a promising source of high-quality training data for future models. Source code is available at https://github.com/gyj155/mssr.

Xiaoxiao Sun, Yunzhong Hou, Hongdong Li et al.

Label-free model evaluation, or AutoEval, estimates model accuracy on unlabeled test sets, and is critical for understanding model behaviors in various unseen environments. In the absence of image labels, based on dataset representations, we estimate model performance for AutoEval with regression. On the one hand, image feature is a straightforward choice for such representations, but it hampers regression learning due to being unstructured (\ie no specific meanings for component at certain location) and of large-scale. On the other hand, previous methods adopt simple structured representations (like average confidence or average feature), but insufficient to capture the data characteristics given their limited dimensions. In this work, we take the best of both worlds and propose a new semi-structured dataset representation that is manageable for regression learning while containing rich information for AutoEval. Based on image features, we integrate distribution shapes, clusters, and representative samples for a semi-structured dataset representation. Besides the structured overall description with distribution shapes, the unstructured description with clusters and representative samples include additional fine-grained information facilitating the AutoEval task. On three existing datasets and 25 newly introduced ones, we experimentally show that the proposed representation achieves competitive results. Code and dataset are available at https://github.com/sxzrt/Semi-Structured-Dataset-Representations.

Xiaomeng Xin, Yiran Zhong, Yunzhong Hou et al.

Regularization-based methods are beneficial to alleviate the catastrophic forgetting problem in class-incremental learning. With the absence of old task images, they often assume that old knowledge is well preserved if the classifier produces similar output on new images. In this paper, we find that their effectiveness largely depends on the nature of old classes: they work well on classes that are easily distinguishable between each other but may fail on more fine-grained ones, e.g., boy and girl. In spirit, such methods project new data onto the feature space spanned by the weight vectors in the fully connected layer, corresponding to old classes. The resulting projections would be similar on fine-grained old classes, and as a consequence the new classifier will gradually lose the discriminative ability on these classes. To address this issue, we propose a memory-free generative replay strategy to preserve the fine-grained old classes characteristics by generating representative old images directly from the old classifier and combined with new data for new classifier training. To solve the homogenization problem of the generated samples, we also propose a diversity loss that maximizes Kullback Leibler (KL) divergence between generated samples. Our method is best complemented by prior regularization-based methods proved to be effective for easily distinguishable old classes. We validate the above design and insights on CUB-200-2011, Caltech-101, CIFAR-100 and Tiny ImageNet and show that our strategy outperforms existing memory-free methods with a clear margin. Code is available at https://github.com/xmengxin/MFGR

Xiaoxiao Sun, Yunzhong Hou, Weijian Deng et al.

Consider a scenario where we are supplied with a number of ready-to-use models trained on a certain source domain and hope to directly apply the most appropriate ones to different target domains based on the models' relative performance. Ideally we should annotate a validation set for model performance assessment on each new target environment, but such annotations are often very expensive. Under this circumstance, we introduce the problem of ranking models in unlabeled new environments. For this problem, we propose to adopt a proxy dataset that 1) is fully labeled and 2) well reflects the true model rankings in a given target environment, and use the performance rankings on the proxy sets as surrogates. We first select labeled datasets as the proxy. Specifically, datasets that are more similar to the unlabeled target domain are found to better preserve the relative performance rankings. Motivated by this, we further propose to search the proxy set by sampling images from various datasets that have similar distributions as the target. We analyze the problem and its solutions on the person re-identification (re-ID) task, for which sufficient datasets are publicly available, and show that a carefully constructed proxy set effectively captures relative performance ranking in new environments. Code is available at \url{https://github.com/sxzrt/Proxy-Set}.

Yunzhong Hou, Liang Zheng

Multiview detection incorporates multiple camera views to deal with occlusions, and its central problem is multiview aggregation. Given feature map projections from multiple views onto a common ground plane, the state-of-the-art method addresses this problem via convolution, which applies the same calculation regardless of object locations. However, such translation-invariant behaviors might not be the best choice, as object features undergo various projection distortions according to their positions and cameras. In this paper, we propose a novel multiview detector, MVDeTr, that adopts a newly introduced shadow transformer to aggregate multiview information. Unlike convolutions, shadow transformer attends differently at different positions and cameras to deal with various shadow-like distortions. We propose an effective training scheme that includes a new view-coherent data augmentation method, which applies random augmentations while maintaining multiview consistency. On two multiview detection benchmarks, we report new state-of-the-art accuracy with the proposed system. Code is available at https://github.com/hou-yz/MVDeTr.

Yunzhong Hou, Liang Zheng

A source model trained on source data and a target model learned through unsupervised domain adaptation (UDA) usually encode different knowledge. To understand the adaptation process, we portray their knowledge difference with image translation. Specifically, we feed a translated image and its original version to the two models respectively, formulating two branches. Through updating the translated image, we force similar outputs from the two branches. When such requirements are met, differences between the two images can compensate for and hence represent the knowledge difference between models. To enforce similar outputs from the two branches and depict the adapted knowledge, we propose a source-free image translation method that generates source-style images using only target images and the two models. We visualize the adapted knowledge on several datasets with different UDA methods and find that generated images successfully capture the style difference between the two domains. For application, we show that generated images enable further tuning of the target model without accessing source data. Code available at https://github.com/hou-yz/DA_visualization.

Yunzhong Hou, Liang Zheng, Stephen Gould

Incorporating multiple camera views for detection alleviates the impact of occlusions in crowded scenes. In a multiview system, we need to answer two important questions when dealing with ambiguities that arise from occlusions. First, how should we aggregate cues from the multiple views? Second, how should we aggregate unreliable 2D and 3D spatial information that has been tainted by occlusions? To address these questions, we propose a novel multiview detection system, MVDet. For multiview aggregation, existing methods combine anchor box features from the image plane, which potentially limits performance due to inaccurate anchor box shapes and sizes. In contrast, we take an anchor-free approach to aggregate multiview information by projecting feature maps onto the ground plane (bird's eye view). To resolve any remaining spatial ambiguity, we apply large kernel convolutions on the ground plane feature map and infer locations from detection peaks. Our entire model is end-to-end learnable and achieves 88.2% MODA on the standard Wildtrack dataset, outperforming the state-of-the-art by 14.1%. We also provide detailed analysis of MVDet on a newly introduced synthetic dataset, MultiviewX, which allows us to control the level of occlusion. Code and MultiviewX dataset are available at https://github.com/hou-yz/MVDet.

Yunzhong Hou, Liang Zheng, Stephen Gould

Color and structure are the two pillars that construct an image. Usually, the structure is well expressed through a rich spectrum of colors, allowing objects in an image to be recognized by neural networks. However, under extreme limitations of color space, the structure tends to vanish, and thus a neural network might fail to understand the image. Interested in exploring this interplay between color and structure, we study the scientific problem of identifying and preserving the most informative image structures while constraining the color space to just a few bits, such that the resulting image can be recognized with possibly high accuracy. To this end, we propose a color quantization network, ColorCNN, which learns to structure the images from the classification loss in an end-to-end manner. Given a color space size, ColorCNN quantizes colors in the original image by generating a color index map and an RGB color palette. Then, this color-quantized image is fed to a pre-trained task network to evaluate its performance. In our experiment, with only a 1-bit color space (i.e., two colors), the proposed network achieves 82.1% top-1 accuracy on the CIFAR10 dataset, outperforming traditional color quantization methods by a large margin. For applications, when encoded with PNG, the proposed color quantization shows superiority over other image compression methods in the extremely low bit-rate regime. The code is available at: https://github.com/hou-yz/color_distillation.

Yunzhong Hou, Liang Zheng, Philip Torr

This study seeks to automate camera movement control for filming existing subjects into attractive videos, contrasting with the creation of non-existent content by directly generating the pixels. We select drone videos as our test case due to their rich and challenging motion patterns, distinctive viewing angles, and precise controls. Existing AI videography methods struggle with limited appearance diversity in simulation training, high costs of recording expert operations, and difficulties in designing heuristic-based goals to cover all scenarios. To avoid these issues, we propose a scalable method that involves collecting real-world training data to improve diversity, extracting camera trajectories automatically to minimize annotation costs, and training an effective architecture that does not rely on heuristics. Specifically, we collect 99k high-quality trajectories by running 3D reconstruction on online videos, connecting camera poses from consecutive frames to formulate 3D camera paths, and using Kalman filter to identify and remove low-quality data. Moreover, we introduce DVGFormer, an auto-regressive transformer that leverages the camera path and images from all past frames to predict camera movement in the next frame. We evaluate our system across 38 synthetic natural scenes and 7 real city 3D scans. We show that our system effectively learns to perform challenging camera movements such as navigating through obstacles, maintaining low altitude to increase perceived speed, and orbiting towers and buildings, which are very useful for recording high-quality videos. Data and code are available at dvgformer.github.io.

Yunzhong Hou, Xingjian Leng, Tom Gedeon et al.

Jointly considering multiple camera views (multi-view) is very effective for pedestrian detection under occlusion. For such multi-view systems, it is critical to have well-designed camera configurations, including camera locations, directions, and fields-of-view (FoVs). Usually, these configurations are crafted based on human experience or heuristics. In this work, we present a novel solution that features a transformer-based camera configuration generator. Using reinforcement learning, this generator autonomously explores vast combinations within the action space and searches for configurations that give the highest detection accuracy according to the training dataset. The generator learns advanced techniques like maximizing coverage, minimizing occlusion, and promoting collaboration. Across multiple simulation scenarios, the configurations generated by our transformer-based model consistently outperform random search, heuristic-based methods, and configurations designed by human experts, shedding light on future camera layout optimization.

Yixin Zhang, Yunzhong Hou, Longqi Li et al.

Ultrasound perception typically requires multiple scan views through probe movement to reduce diagnostic ambiguity, mitigate acoustic occlusions, and improve anatomical coverage. However, not all probe views are equally informative. Exhaustively acquiring a large number of views can introduce substantial redundancy, increase scanning and processing costs. To address this, we define an active view exploration task for ultrasound and propose SonoSelect, an ultrasound-specific method that adaptively guides probe movement based on current observations. Specifically, we cast ultrasound active view exploration as a sequential decision-making problem. Each new 2D ultrasound view is fused into a 3D spatial memory of the observed anatomy, which guides the next probe position. On top of this formulation, we propose an ultrasound-specific objective that favors probe movements with greater organ coverage, lower reconstruction uncertainty, and less redundant scanning. Experiments on the ultrasound simulator show that SonoSelect achieves promising multi-view organ classification accuracy using only 2 out of N views. Furthermore, for a more difficult kidney cyst detection task, it reaches 54.56% kidney coverage and 35.13% cyst coverage, with short trajectories consistently centered on the target cyst.

Changlin Song, Yunzhong Hou, Michael Randall Barnes et al.

Extreme amodal detection is the task of inferring the 2D location of objects that are not fully visible in the input image but are visible within an expanded field-of-view. This differs from amodal detection, where the object is partially visible within the input image, but is occluded. In this paper, we consider the sub-problem of face detection, since this class provides motivating applications involving safety and privacy, but do not tailor our method specifically to this class. Existing approaches rely on image sequences so that missing detections may be interpolated from surrounding frames or make use of generative models to sample possible completions. In contrast, we consider the single-image task and propose a more efficient, sample-free approach that makes use of the contextual cues from the image to infer the presence of unseen faces. We design a heatmap-based extreme amodal object detector that addresses the problem of efficiently predicting a lot (the out-of-frame region) from a little (the image) with a selective coarse-to-fine decoder. Our method establishes strong results for this new task, even outperforming less efficient generative approaches.

Yunzhong Hou, Zhongdao Wang, Shengjin Wang et al.

Data associations in multi-target multi-camera tracking (MTMCT) usually estimate affinity directly from re-identification (re-ID) feature distances. However, we argue that it might not be the best choice given the difference in matching scopes between re-ID and MTMCT problems. Re-ID systems focus on global matching, which retrieves targets from all cameras and all times. In contrast, data association in tracking is a local matching problem, since its candidates only come from neighboring locations and time frames. In this paper, we design experiments to verify such misfit between global re-ID feature distances and local matching in tracking, and propose a simple yet effective approach to adapt affinity estimations to corresponding matching scopes in MTMCT. Instead of trying to deal with all appearance changes, we tailor the affinity metric to specialize in ones that might emerge during data associations. To this end, we introduce a new data sampling scheme with temporal windows originally used for data associations in tracking. Minimizing the mismatch, the adaptive affinity module brings significant improvements over global re-ID distance, and produces competitive performance on CityFlow and DukeMTMC datasets.

Yunzhong Hou, Liang Zheng

Effort in releasing large-scale datasets may be compromised by privacy and intellectual property considerations. A feasible alternative is to release pre-trained models instead. While these models are strong on their original task (source domain), their performance might degrade significantly when deployed directly in a new environment (target domain), which might not contain labels for training under realistic settings. Domain adaptation (DA) is a known solution to the domain gap problem, but usually requires labeled source data. In this paper, we study the problem of source free domain adaptation (SFDA), whose distinctive feature is that the source domain only provides a pre-trained model, but no source data. Being source free adds significant challenges to DA, especially when considering that the target dataset is unlabeled. To solve the SFDA problem, we propose an image translation approach that transfers the style of target images to that of unseen source images. To this end, we align the batch-wise feature statistics of generated images to that stored in batch normalization layers of the pre-trained model. Compared with directly classifying target images, higher accuracy is obtained with these style transferred images using the pre-trained model. On several image classification datasets, we show that the above-mentioned improvements are consistent and statistically significant.

Yunzhong Hou, Liang Zheng, Zhongdao Wang et al.

Multi-target multi-camera tracking (MTMCT) systems track targets across cameras. Due to the continuity of target trajectories, tracking systems usually restrict their data association within a local neighborhood. In single camera tracking, local neighborhood refers to consecutive frames; in multi-camera tracking, it refers to neighboring cameras that the target may appear successively. For similarity estimation, tracking systems often adopt appearance features learned from the re-identification (re-ID) perspective. Different from tracking, re-ID usually does not have access to the trajectory cues that can limit the search space to a local neighborhood. Due to its global matching property, the re-ID perspective requires to learn global appearance features. We argue that the mismatch between the local matching procedure in tracking and the global nature of re-ID appearance features may compromise MTMCT performance. To fit the local matching procedure in MTMCT, in this work, we introduce locality aware appearance metric (LAAM). Specifically, we design an intra-camera metric for single camera tracking, and an inter-camera metric for multi-camera tracking. Both metrics are trained with data pairs sampled from their corresponding local neighborhoods, as opposed to global sampling in the re-ID perspective. We show that the locally learned metrics can be successfully applied on top of several globally learned re-ID features. With the proposed method, we report new state-of-the-art performance on the DukeMTMC dataset, and a substantial improvement on the CityFlow dataset.

Wenqi Shi, Yunzhong Hou, Sheng Zhou et al.

Deep neural networks (DNNs) are state-of-the-art solutions for many machine learning applications, and have been widely used on mobile devices. Running DNNs on resource-constrained mobile devices often requires the help from edge servers via computation offloading. However, offloading through a bandwidth-limited wireless link is non-trivial due to the tight interplay between the computation resources on mobile devices and wireless resources. Existing studies have focused on cooperative inference where DNN models are partitioned at different neural network layers, and the two parts are executed at the mobile device and the edge server, respectively. Since the output data size of a DNN layer can be larger than that of the raw data, offloading intermediate data between layers can suffer from high transmission latency under limited wireless bandwidth. In this paper, we propose an efficient and flexible 2-step pruning framework for DNN partition between mobile devices and edge servers. In our framework, the DNN model only needs to be pruned once in the training phase where unimportant convolutional filters are removed iteratively. By limiting the pruning region, our framework can greatly reduce either the wireless transmission workload of the device or the total computation workload. A series of pruned models are generated in the training phase, from which the framework can automatically select to satisfy varying latency and accuracy requirements. Furthermore, coding for the intermediate data is added to provide extra transmission workload reduction. Our experiments show that the proposed framework can achieve up to 25.6$\times$ reduction on transmission workload, 6.01$\times$ acceleration on total computation and 4.81$\times$ reduction on end-to-end latency as compared to partitioning the original DNN model without pruning.