Haoxiang Li

Haichao Yu, Haoxiang Li, Gang Hua et al. · gatech, tsinghua

Early-exiting dynamic neural networks (EDNN), as one type of dynamic neural networks, has been widely studied recently. A typical EDNN has multiple prediction heads at different layers of the network backbone. During inference, the model will exit at either the last prediction head or an intermediate prediction head where the prediction confidence is higher than a predefined threshold. To optimize the model, these prediction heads together with the network backbone are trained on every batch of training data. This brings a train-test mismatch problem that all the prediction heads are optimized on all types of data in training phase while the deeper heads will only see difficult inputs in testing phase. Treating training and testing inputs differently at the two phases will cause the mismatch between training and testing data distributions. To mitigate this problem, we formulate an EDNN as an additive model inspired by gradient boosting, and propose multiple training techniques to optimize the model effectively. We name our method BoostNet. Our experiments show it achieves the state-of-the-art performance on CIFAR100 and ImageNet datasets in both anytime and budgeted-batch prediction modes. Our code is released at https://github.com/SHI-Labs/Boosted-Dynamic-Networks.

Biqing Qi, Kaiyan Zhang, Haoxiang Li et al.

Significant scientific discoveries have driven the progress of human civilisation. The explosion of scientific literature and data has created information barriers across disciplines that have slowed the pace of scientific discovery. Large Language Models (LLMs) hold a wealth of global and interdisciplinary knowledge that promises to break down these information barriers and foster a new wave of scientific discovery. However, the potential of LLMs for scientific discovery has not been formally explored. In this paper, we start from investigating whether LLMs can propose scientific hypotheses. To this end, we construct a dataset consist of background knowledge and hypothesis pairs from biomedical literature. The dataset is divided into training, seen, and unseen test sets based on the publication date to control visibility. We subsequently evaluate the hypothesis generation capabilities of various top-tier instructed models in zero-shot, few-shot, and fine-tuning settings, including both closed and open-source LLMs. Additionally, we introduce an LLM-based multi-agent cooperative framework with different role designs and external tools to enhance the capabilities related to generating hypotheses. We also design four metrics through a comprehensive review to evaluate the generated hypotheses for both ChatGPT-based and human evaluations. Through experiments and analyses, we arrive at the following findings: 1) LLMs surprisingly generate untrained yet validated hypotheses from testing literature. 2) Increasing uncertainty facilitates candidate generation, potentially enhancing zero-shot hypothesis generation capabilities. These findings strongly support the potential of LLMs as catalysts for new scientific discoveries and guide further exploration.

Jiaxin Lu, Hao Kang, Haoxiang Li et al. · bytedance

Dexterous grasping aims to produce diverse grasping postures with a high grasping success rate. Regression-based methods that directly predict grasping parameters given the object may achieve a high success rate but often lack diversity. Generation-based methods that generate grasping postures conditioned on the object can often produce diverse grasping, but they are insufficient for high grasping success due to lack of discriminative information. To mitigate, we introduce a unified diffusion-based dexterous grasp generation model, dubbed the name UGG, which operates within the object point cloud and hand parameter spaces. Our all-transformer architecture unifies the information from the object, the hand, and the contacts, introducing a novel representation of contact points for improved contact modeling. The flexibility and quality of our model enable the integration of a lightweight discriminator, benefiting from simulated discriminative data, which pushes for a high success rate while preserving high diversity. Beyond grasp generation, our model can also generate objects based on hand information, offering valuable insights into object design and studying how the generative model perceives objects. Our model achieves state-of-the-art dexterous grasping on the large-scale DexGraspNet dataset while facilitating human-centric object design, marking a significant advancement in dexterous grasping research. Our project page is https://jiaxin-lu.github.io/ugg/.

Siwei Li, Mingxuan Liu, Yating Zhang et al. · tsinghua

Image deblurring is a critical task in the field of image restoration, aiming to eliminate blurring artifacts. However, the challenge of addressing non-uniform blurring leads to an ill-posed problem, which limits the generalization performance of existing deblurring models. To solve the problem, we propose a framework SAM-Deblur, integrating prior knowledge from the Segment Anything Model (SAM) into the deblurring task for the first time. In particular, SAM-Deblur is divided into three stages. First, we preprocess the blurred images, obtain segment masks via SAM, and propose a mask dropout method for training to enhance model robustness. Then, to fully leverage the structural priors generated by SAM, we propose a Mask Average Pooling (MAP) unit specifically designed to average SAM-generated segmented areas, serving as a plug-and-play component which can be seamlessly integrated into existing deblurring networks. Finally, we feed the fused features generated by the MAP Unit into the deblurring model to obtain a sharp image. Experimental results on the RealBlurJ, ReloBlur, and REDS datasets reveal that incorporating our methods improves GoPro-trained NAFNet's PSNR by 0.05, 0.96, and 7.03, respectively. Project page is available at GitHub \href{https://hplqaq.github.io/projects/sam-deblur}{HPLQAQ/SAM-Deblur}.

Biqing Qi, Kaiyan Zhang, Kai Tian et al.

The rapid growth of biomedical knowledge has outpaced our ability to efficiently extract insights and generate novel hypotheses. Large language models (LLMs) have emerged as a promising tool to revolutionize knowledge interaction and potentially accelerate biomedical discovery. In this paper, we present a comprehensive evaluation of LLMs as biomedical hypothesis generators. We construct a dataset of background-hypothesis pairs from biomedical literature, carefully partitioned into training, seen, and unseen test sets based on publication date to mitigate data contamination. Using this dataset, we assess the hypothesis generation capabilities of top-tier instructed models in zero-shot, few-shot, and fine-tuning settings. To enhance the exploration of uncertainty, a crucial aspect of scientific discovery, we incorporate tool use and multi-agent interactions in our evaluation framework. Furthermore, we propose four novel metrics grounded in extensive literature review to evaluate the quality of generated hypotheses, considering both LLM-based and human assessments. Our experiments yield two key findings: 1) LLMs can generate novel and validated hypotheses, even when tested on literature unseen during training, and 2) Increasing uncertainty through multi-agent interactions and tool use can facilitate diverse candidate generation and improve zero-shot hypothesis generation performance. However, we also observe that the integration of additional knowledge through few-shot learning and tool use may not always lead to performance gains, highlighting the need for careful consideration of the type and scope of external knowledge incorporated. These findings underscore the potential of LLMs as powerful aids in biomedical hypothesis generation and provide valuable insights to guide further research in this area.

Zheng Chong, Xiao Dong, Haoxiang Li et al.

Virtual try-on methods based on diffusion models achieve realistic effects but often require additional encoding modules, a large number of training parameters, and complex preprocessing, which increases the burden on training and inference. In this work, we re-evaluate the necessity of additional modules and analyze how to improve training efficiency and reduce redundant steps in the inference process. Based on these insights, we propose CatVTON, a simple and efficient virtual try-on diffusion model that transfers in-shop or worn garments of arbitrary categories to target individuals by concatenating them along spatial dimensions as inputs of the diffusion model. The efficiency of CatVTON is reflected in three aspects: (1) Lightweight network. CatVTON consists only of a VAE and a simplified denoising UNet, removing redundant image and text encoders as well as cross-attentions, and includes just 899.06M parameters. (2) Parameter-efficient training. Through experimental analysis, we identify self-attention modules as crucial for adapting pre-trained diffusion models to the virtual try-on task, enabling high-quality results with only 49.57M training parameters. (3) Simplified inference. CatVTON eliminates unnecessary preprocessing, such as pose estimation, human parsing, and captioning, requiring only a person image and garment reference to guide the virtual try-on process, reducing over 49% memory usage compared to other diffusion-based methods. Extensive experiments demonstrate that CatVTON achieves superior qualitative and quantitative results compared to baseline methods and demonstrates strong generalization performance in in-the-wild scenarios, despite being trained solely on public datasets with 73K samples.

Lei Fan, Bo Liu, Haoxiang Li et al.

In real-world scenarios, typical visual recognition systems could fail under two major causes, i.e., the misclassification between known classes and the excusable misbehavior on unknown-class images. To tackle these deficiencies, flexible visual recognition should dynamically predict multiple classes when they are unconfident between choices and reject making predictions when the input is entirely out of the training distribution. Two challenges emerge along with this novel task. First, prediction uncertainty should be separately quantified as confusion depicting inter-class uncertainties and ignorance identifying out-of-distribution samples. Second, both confusion and ignorance should be comparable between samples to enable effective decision-making. In this paper, we propose to model these two sources of uncertainty explicitly with the theory of Subjective Logic. Regarding recognition as an evidence-collecting process, confusion is then defined as conflicting evidence, while ignorance is the absence of evidence. By predicting Dirichlet concentration parameters for singletons, comprehensive subjective opinions, including confusion and ignorance, could be achieved via further evidence combinations. Through a series of experiments on synthetic data analysis, visual recognition, and open-set detection, we demonstrate the effectiveness of our methods in quantifying two sources of uncertainties and dealing with flexible recognition.

Xiangyu Xu, Li Guan, Enrique Dunn et al.

In this paper, we propose an end-to-end framework that jointly learns keypoint detection, descriptor representation and cross-frame matching for the task of image-based 3D localization. Prior art has tackled each of these components individually, purportedly aiming to alleviate difficulties in effectively train a holistic network. We design a self-supervised image warping correspondence loss for both feature detection and matching, a weakly-supervised epipolar constraints loss on relative camera pose learning, and a directional matching scheme that detects key-point features in a source image and performs coarse-to-fine correspondence search on the target image. We leverage this framework to enforce cycle consistency in our matching module. In addition, we propose a new loss to robustly handle both definite inlier/outlier matches and less-certain matches. The integration of these learning mechanisms enables end-to-end training of a single network performing all three localization components. Bench-marking our approach on public data-sets, exemplifies how such an end-to-end framework is able to yield more accurate localization that out-performs both traditional methods as well as state-of-the-art weakly supervised methods.

Zeyang Liu, Le Wang, Sanping Zhou et al.

Content-aware layout generation is a critical task in graphic design automation, focused on creating visually appealing arrangements of elements that seamlessly blend with a given background image. The variety of real-world applications makes it highly challenging to develop a single model capable of unifying the diverse range of input-constrained generation sub-tasks, such as those conditioned by element types, sizes, or their relationships. Current methods either address only a subset of these tasks or necessitate separate model parameters for different conditions, failing to offer a truly unified solution. In this paper, we propose UniLayDiff: a Unified Diffusion Transformer, that for the first time, addresses various content-aware layout generation tasks with a single, end-to-end trainable model. Specifically, we treat layout constraints as a distinct modality and employ Multi-Modal Diffusion Transformer framework to capture the complex interplay between the background image, layout elements, and diverse constraints. Moreover, we integrate relation constraints through fine-tuning the model with LoRA after pretraining the model on other tasks. Such a schema not only achieves unified conditional generation but also enhances overall layout quality. Extensive experiments demonstrate that UniLayDiff achieves state-of-the-art performance across from unconditional to various conditional generation tasks and, to the best of our knowledge, is the first model to unify the full range of content-aware layout generation tasks.

Yuyang Sha, Zijie Lou, Youyun Tang et al.

Portrait composition plays a central role in portrait aesthetics and visual communication, yet existing datasets and benchmarks mainly focus on coarse aesthetic scoring, generic image aesthetics, or unconstrained portrait generation. This limits systematic research on structured portrait composition analysis and controllable portrait generation under explicit composition requirements. In this paper, we introduce PortraitCraft, a unified benchmark for portrait composition understanding and generation. PortraitCraft is built on a dataset of approximately 50,000 curated real portrait images with structured multi-level supervision, including global composition scores, annotations over 13 composition attributes, attribute-level explanation texts, visual question answering pairs, and composition-oriented textual descriptions for generation. Based on this dataset, we establish two complementary benchmark tasks for composition understanding and composition-aware generation within a unified framework. The first evaluates portrait composition understanding through score prediction, fine-grained attribute reasoning, and image-grounded visual question answering, while the second evaluates portrait generation from structured composition descriptions under explicit composition constraints. We further define standardized evaluation protocols and provide reference baseline results with representative multimodal models. PortraitCraft provides a comprehensive benchmark for future research on fine-grained portrait understanding, interpretable aesthetic assessment, and controllable portrait generation.

Xiaotian Hu, Mingxuan Liu, Hongjia Yang et al.

Spatio-temporal fetal brain atlases are important for characterizing normative neurodevelopment and identifying congenital anomalies. However, existing atlas construction pipelines necessitate days for slice-to-volume reconstruction (SVR) to generate high-resolution 3D brain volumes and several additional days for iterative volume registration, thereby rendering atlas construction from large-scale cohorts prohibitively impractical. We address these limitations with INFANiTE, an Implicit Neural Representation (INR) framework for high-resolution Fetal brain spatio-temporal Atlas learNing from clinical Thick-slicE MRI scans, bypassing both the costly SVR and the iterative non-rigid registration steps entirely, thereby substantially accelerating atlas construction. Extensive experiments demonstrate that INFANiTE outperforms existing baselines in subject consistency, reference fidelity, intrinsic quality and biological plausibility, even under challenging sparse-data settings. Additionally, INFANiTE reduces the end-to-end processing time (i.e., from raw scans to the final atlas) from days to hours compared to the traditional 3D volume-based pipeline (e.g., SyGN), facilitating large-scale population-level fetal brain analysis. Our code is publicly available at: https://anonymous.4open.science/r/INFANiTE-5D74

Zixin Zhu, Haoxiang Li, Xuelu Feng et al.

Towards intelligent image editing, object removal should eliminate both the target object and its causal visual artifacts, such as shadows and reflections. However, existing image appearance-based methods either follow strictly mask-aligned training and fail to remove these causal effects which are not explicitly masked, or adopt loosely mask-aligned strategies that lack controllability and may unintentionally over-erase other objects. We identify that these limitations stem from ignoring the causal relationship between an object's geometry presence and its visual effects. To address this limitation, we propose a geometry-aware two-stage framework that decouples object removal into (1) geometry removal and (2) appearance rendering. In the first stage, we remove the object directly from the geometry (e.g., depth) using strictly mask-aligned supervision, enabling structure-aware editing with strong geometric constraints. In the second stage, we render a photorealistic RGB image conditioned on the updated geometry, where causal visual effects are considered implicitly as a result of the modified 3D geometry. To guide learning in the geometry removal stage, we introduce a preference-driven objective based on positive and negative sample pairs, encouraging the model to remove objects as well as their causal visual artifacts while avoiding new structural insertions. Extensive experiments demonstrate that our method achieves state-of-the-art performance in removing both objects and their associated artifacts on two popular benchmarks. The code is available at https://github.com/buxiangzhiren/GeoRemover.

Kumara Kahatapitiya, Zhou Ren, Haoxiang Li et al.

Temporal Activity Detection aims to predict activity classes per frame, in contrast to video-level predictions in Activity Classification (i.e., Activity Recognition). Due to the expensive frame-level annotations required for detection, the scale of detection datasets is limited. Thus, commonly, previous work on temporal activity detection resorts to fine-tuning a classification model pretrained on large-scale classification datasets (e.g., Kinetics-400). However, such pretrained models are not ideal for downstream detection, due to the disparity between the pretraining and the downstream fine-tuning tasks. In this work, we propose a novel 'weakly-guided self-supervised' pretraining method for detection. We leverage weak labels (classification) to introduce a self-supervised pretext task (detection) by generating frame-level pseudo labels, multi-action frames, and action segments. Simply put, we design a detection task similar to downstream, on large-scale classification data, without extra annotations. We show that the models pretrained with the proposed weakly-guided self-supervised detection task outperform prior work on multiple challenging activity detection benchmarks, including Charades and MultiTHUMOS. Our extensive ablations further provide insights on when and how to use the proposed models for activity detection. Code is available at https://github.com/kkahatapitiya/SSDet.

Haichao Yu, Haoxiang Li, Honghui Shi et al.

We present any-precision deep neural networks (DNNs), which are trained with a new method that allows the learned DNNs to be flexible in numerical precision during inference. The same model in runtime can be flexibly and directly set to different bit-widths, by truncating the least significant bits, to support dynamic speed and accuracy trade-off. When all layers are set to low-bits, we show that the model achieved accuracy comparable to dedicated models trained at the same precision. This nice property facilitates flexible deployment of deep learning models in real-world applications, where in practice trade-offs between model accuracy and runtime efficiency are often sought. Previous literature presents solutions to train models at each individual fixed efficiency/accuracy trade-off point. But how to produce a model flexible in runtime precision is largely unexplored. When the demand of efficiency/accuracy trade-off varies from time to time or even dynamically changes in runtime, it is infeasible to re-train models accordingly, and the storage budget may forbid keeping multiple models. Our proposed framework achieves this flexibility without performance degradation. More importantly, we demonstrate that this achievement is agnostic to model architectures and applicable to multiple vision tasks. Our code is released at https://github.com/SHI-Labs/Any-Precision-DNNs.

Mingxuan Liu, Yingqi Hao, Yi Liao et al.

Background: Prenatal germinal matrix-intraventricular hemorrhage (GMH-IVH) is a leading cause of infant mortality and neurodevelopmental impairment. Manual diagnosis and lesion segmentation are labor-intensive and error-prone. Deep learning models offer potential for automation but typically require large annotated datasets, which are challenging to obtain. Purpose: To develop and validate an annotation-free deep learning framework for automated detection and segmentation of GMH-IVH on brain MRI. Materials and Methods: This retrospective study analyzed 2D T2-weighted MRI data from pregnant women collected from October 2015 to October 2023 at one hospital (internal validation) and two hospitals (external validation). Eligible participants included healthy fetuses and those with GMH-IVH. FreeHemoSeg was developed and trained using pseudo GMH-IVH images synthesized from normal fetal data guided by medical priors. Primary outcomes included diagnostic accuracy (area under the ROC curve [AUROC], sensitivity, specificity) and segmentation accuracy (Dice similarity coefficient [DSC]). A reader study evaluated clinical utility. Results: A total of 1674 stacks from 558 pregnant women were analyzed. FreeHemoSeg achieved the highest performance in both internal (sensitivity: 0.914, 95% CI 0.869-0.945; specificity: 0.966, 95% CI 0.946-0.978; DSC: 0.559, 95% CI 0.546-0.571) and external validation (sensitivity: 0.824, 95% CI 0.739-0.885; specificity: 0.943, 95% CI 0.913-0.964; DSC: 0.512, 95% CI 0.497-0.526), outperforming supervised and unsupervised methods. FreeHemoSeg assistance improved radiologists' sensitivity (from 0.882 to 0.941-1.000) and diagnostic confidence while reducing interpretation time by 16.0-52.7%. Conclusion: FreeHemoSeg accurately detects and localizes fetal brain hemorrhages without annotated training data, enabling earlier diagnosis and supporting timely clinical management.

Adel ElZemity, Budi Arief, Shujun Li et al.

Bare-metal operational technology (OT) devices -- especially the microcontrollers running Modbus/TCP and CoAP at the base of industrial control systems -- have remained outside the reach of autonomous security attacks. Prior autonomous pentesting studies target Linux and web systems, whose shells and filesystems are familiar to LLM agents. Bare-metal OT has neither, so agents must reason directly over protocol fields and parser semantics. This requires new action-space designs and runtime controls, and opens new research questions about protocol-level exploit reasoning and its deployment envelope. We present APIOT (Autonomous Purple-teaming for Industrial OT), the first large language model (LLM) framework demonstrating an autonomous attack and remediation of bare-metal OT devices, achieving the full discovery -> exploitation -> patching -> verification cycle without step-by-step human intervention. We implemented and evaluated this framework on Zephyr RTOS firmware across heterogeneous industrial IoT (IIoT) topologies. Through 290 experiment runs spanning five frontier LLMs, three network topologies, two impairment levels, and guided versus unguided conditions, APIOT achieved a mission success rate of 90.0% on the full attack-remediation cycle. We found that the runtime governance layer (which we call an overseer) is a critical engineering variable: without it, agents exhibit systematic degenerate patterns, including repetition loops, missing crash verification, and reconnaissance deadlocks. Together, these findings carry two implications beyond our testbed. Attacker expertise is no longer the binding constraint on bare-metal OT exploitation, and defender threat models must now assume LLM-augmented adversaries capable of executing autonomous discovery-through-remediation cycles against industrial firmware.

Mingxuan Liu, Jiankai Tang, Yongli Chen et al. · tsinghua

Artificial neural networks (ANNs) can help camera-based remote photoplethysmography (rPPG) in measuring cardiac activity and physiological signals from facial videos, such as pulse wave, heart rate and respiration rate with better accuracy. However, most existing ANN-based methods require substantial computing resources, which poses challenges for effective deployment on mobile devices. Spiking neural networks (SNNs), on the other hand, hold immense potential for energy-efficient deep learning owing to their binary and event-driven architecture. To the best of our knowledge, we are the first to introduce SNNs into the realm of rPPG, proposing a hybrid neural network (HNN) model, the Spiking-PhysFormer, aimed at reducing power consumption. Specifically, the proposed Spiking-PhyFormer consists of an ANN-based patch embedding block, SNN-based transformer blocks, and an ANN-based predictor head. First, to simplify the transformer block while preserving its capacity to aggregate local and global spatio-temporal features, we design a parallel spike transformer block to replace sequential sub-blocks. Additionally, we propose a simplified spiking self-attention mechanism that omits the value parameter without compromising the model's performance. Experiments conducted on four datasets-PURE, UBFC-rPPG, UBFC-Phys, and MMPD demonstrate that the proposed model achieves a 12.4\% reduction in power consumption compared to PhysFormer. Additionally, the power consumption of the transformer block is reduced by a factor of 12.2, while maintaining decent performance as PhysFormer and other ANN-based models.

Xiangrui Liu, Haoxiang Li, Yezhou Yang

Video relighting offers immense creative potential and commercial value but is hindered by challenges, including the absence of an adequate evaluation metric, severe light flickering, and the degradation of fine-grained details during editing. To overcome these challenges, we introduce Hi-Light, a novel, training-free framework for high-fidelity, high-resolution, robust video relighting. Our approach introduces three technical innovations: lightness prior anchored guided relighting diffusion that stabilises intermediate relit video, a Hybrid Motion-Adaptive Lighting Smoothing Filter that leverages optical flow to ensure temporal stability without introducing motion blur, and a LAB-based Detail Fusion module that preserves high-frequency detail information from the original video. Furthermore, to address the critical gap in evaluation, we propose the Light Stability Score, the first quantitative metric designed to specifically measure lighting consistency. Extensive experiments demonstrate that Hi-Light significantly outperforms state-of-the-art methods in both qualitative and quantitative comparisons, producing stable, highly detailed relit videos.

Zheng Chong, Wenqing Zhang, Shiyue Zhang et al.

Virtual try-on (VTON) technology has gained attention due to its potential to transform online retail by enabling realistic clothing visualization of images and videos. However, most existing methods struggle to achieve high-quality results across image and video try-on tasks, especially in long video scenarios. In this work, we introduce CatV2TON, a simple and effective vision-based virtual try-on (V2TON) method that supports both image and video try-on tasks with a single diffusion transformer model. By temporally concatenating garment and person inputs and training on a mix of image and video datasets, CatV2TON achieves robust try-on performance across static and dynamic settings. For efficient long-video generation, we propose an overlapping clip-based inference strategy that uses sequential frame guidance and Adaptive Clip Normalization (AdaCN) to maintain temporal consistency with reduced resource demands. We also present ViViD-S, a refined video try-on dataset, achieved by filtering back-facing frames and applying 3D mask smoothing for enhanced temporal consistency. Comprehensive experiments demonstrate that CatV2TON outperforms existing methods in both image and video try-on tasks, offering a versatile and reliable solution for realistic virtual try-ons across diverse scenarios.

Yunpeng Bai, Haoxiang Li, Qixing Huang

Diffusion Transformers (DiTs) have emerged as the dominant architecture for visual generation, powering state-of-the-art image and video models. By representing images as patch tokens with positional encodings (PEs), DiTs combine Transformer scalability with spatial and temporal inductive biases. In this work, we revisit how DiTs organize visual content and discover that patch tokens exhibit a surprising degree of independence: even when PEs are perturbed, DiTs still produce globally coherent outputs, indicating that spatial coherence is primarily governed by PEs. Motivated by this finding, we introduce the Positional Encoding Field (PE-Field), which extends positional encodings from the 2D plane to a structured 3D field. PE-Field incorporates depth-aware encodings for volumetric reasoning and hierarchical encodings for fine-grained sub-patch control, enabling DiTs to model geometry directly in 3D space. Our PE-Field-augmented DiT achieves state-of-the-art performance on single-image novel view synthesis and generalizes to controllable spatial image editing.

Yuxuan Wu, Le Wang, Sanping Zhou et al.

Controllable layout generation aims to create plausible visual arrangements of element bounding boxes within a graphic design according to certain optional constraints, such as the type or position of a specific component. While recent diffusion or flow-matching models have achieved considerable advances in multifarious conditional generation tasks, there remains considerable room for generating optimal arrangements under given conditions. In this work, we propose to carry out layout generation through retrieving by conditions and reference-guided generation. Specifically, we retrieve appropriate layout templates according to given conditions as references. The references are then utilized to guide the denoising or flow-based transport process. By retrieving layouts compatible with the given conditions, we can uncover the potential information not explicitly provided in the given condition. Such an approach offers more effective guidance to the model during the generation process, in contrast to previous models that feed the condition to the model and let the model infer the unprovided layout attributes directly. Meanwhile, we design a condition-modulated attention that selectively absorbs retrieval knowledge, adapting to the difference between retrieved templates and given conditions. Extensive experiment results show that our method successfully produces high-quality layouts that meet the given conditions and outperforms existing state-of-the-art models. Code will be released upon acceptance.

Bo Sun, Hao Kang, Li Guan et al.

We present a deep learning model, dubbed Glissando-Net, to simultaneously estimate the pose and reconstruct the 3D shape of objects at the category level from a single RGB image. Previous works predominantly focused on either estimating poses(often at the instance level), or reconstructing shapes, but not both. Glissando-Net is composed of two auto-encoders that are jointly trained, one for RGB images and the other for point clouds. We embrace two key design choices in Glissando-Net to achieve a more accurate prediction of the 3D shape and pose of the object given a single RGB image as input. First, we augment the feature maps of the point cloud encoder and decoder with transformed feature maps from the image decoder, enabling effective 2D-3D interaction in both training and prediction. Second, we predict both the 3D shape and pose of the object in the decoder stage. This way, we better utilize the information in the 3D point clouds presented only in the training stage to train the network for more accurate prediction. We jointly train the two encoder-decoders for RGB and point cloud data to learn how to pass latent features to the point cloud decoder during inference. In testing, the encoder of the 3D point cloud is discarded. The design of Glissando-Net is inspired by codeSLAM. Unlike codeSLAM, which targets 3D reconstruction of scenes, we focus on pose estimation and shape reconstruction of objects, and directly predict the object pose and a pose invariant 3D reconstruction without the need of the code optimization step. Extensive experiments, involving both ablation studies and comparison with competing methods, demonstrate the efficacy of our proposed method, and compare favorably with the state-of-the-art.

Weikai Lin, Haoxiang Li, Yuhao Zhu

Hair simulation and rendering are challenging due to complex strand dynamics, diverse material properties, and intricate light-hair interactions. Recent video diffusion models can generate high-quality videos, but they lack fine-grained control over hair dynamics. We present ControlHair, a hybrid framework that integrates a physics simulator with conditional video diffusion to enable controllable dynamic hair rendering. ControlHair adopts a three-stage pipeline: it first encodes physics parameters (e.g., hair stiffness, wind) into per-frame geometry using a simulator, then extracts per-frame control signals, and finally feeds control signals into a video diffusion model to generate videos with desired hair dynamics. This cascaded design decouples physics reasoning from video generation, supports diverse physics, and makes training the video diffusion model easy. Trained on a curated 10K video dataset, ControlHair outperforms text- and pose-conditioned baselines, delivering precisely controlled hair dynamics. We further demonstrate three use cases of ControlHair: dynamic hairstyle try-on, bullet-time effects, and cinemagraphic. ControlHair introduces the first physics-informed video diffusion framework for controllable dynamics. We provide a teaser video and experimental results on our website.

Shiyue Zhang, Zheng Chong, Xi Lu et al.

Building on the success of diffusion models, significant advancements have been made in multimodal image generation tasks. Among these, human image generation has emerged as a promising technique, offering the potential to revolutionize the fashion design process. However, existing methods often focus solely on text-to-image or image reference-based human generation, which fails to satisfy the increasingly sophisticated demands. To address the limitations of flexibility and precision in human generation, we introduce ComposeAnyone, a controllable layout-to-human generation method with decoupled multimodal conditions. Specifically, our method allows decoupled control of any part in hand-drawn human layouts using text or reference images, seamlessly integrating them during the generation process. The hand-drawn layout, which utilizes color-blocked geometric shapes such as ellipses and rectangles, can be easily drawn, offering a more flexible and accessible way to define spatial layouts. Additionally, we introduce the ComposeHuman dataset, which provides decoupled text and reference image annotations for different components of each human image, enabling broader applications in human image generation tasks. Extensive experiments on multiple datasets demonstrate that ComposeAnyone generates human images with better alignment to given layouts, text descriptions, and reference images, showcasing its multi-task capability and controllability.

Ziyu Li, Zihan Li, Haoxiang Li et al.

This chapter provides an overview of deep learning techniques for improving the spatial resolution of MRI, ranging from convolutional neural networks, generative adversarial networks, to more advanced models including transformers, diffusion models, and implicit neural representations. Our exploration extends beyond the methodologies to scrutinize the impact of super-resolved images on clinical and neuroscientific assessments. We also cover various practical topics such as network architectures, image evaluation metrics, network loss functions, and training data specifics, including downsampling methods for simulating low-resolution images and dataset selection. Finally, we discuss existing challenges and potential future directions regarding the feasibility and reliability of deep learning-based MRI super-resolution, with the aim to facilitate its wider adoption to benefit various clinical and neuroscientific applications.

Hongjia Yang, Guanhua Wang, Ziyu Li et al.

Magnetic resonance imaging (MRI) has significantly benefited from the resurgence of artificial intelligence (AI). By leveraging AI's capabilities in large-scale optimization and pattern recognition, innovative methods are transforming the MRI acquisition workflow, including planning, sequence design, and correction of acquisition artifacts. These emerging algorithms demonstrate substantial potential in enhancing the efficiency and throughput of acquisition steps. This review discusses several pivotal AI-based methods in neuro MRI acquisition, focusing on their technological advances, impact on clinical practice, and potential risks.

Mo Zhou, Yiding Yang, Haoxiang Li et al. · bytedance

With a strong alignment between the training and test distributions, object relation as a context prior facilitates object detection. Yet, it turns into a harmful but inevitable training set bias upon test distributions that shift differently across space and time. Nevertheless, the existing detectors cannot incorporate deployment context prior during the test phase without parameter update. Such kind of capability requires the model to explicitly learn disentangled representations with respect to context prior. To achieve this, we introduce an additional graph input to the detector, where the graph represents the deployment context prior, and its edge values represent object relations. Then, the detector behavior is trained to bound to the graph with a modified training objective. As a result, during the test phase, any suitable deployment context prior can be injected into the detector via graph edits, hence calibrating, or "re-biasing" the detector towards the given prior at run-time without parameter update. Even if the deployment prior is unknown, the detector can self-calibrate using deployment prior approximated using its own predictions. Comprehensive experimental results on the COCO dataset, as well as cross-dataset testing on the Objects365 dataset, demonstrate the effectiveness of the run-time calibratable detector.

Siming Yan, Zhenpei Yang, Haoxiang Li et al.

This paper advocates the use of implicit surface representation in autoencoder-based self-supervised 3D representation learning. The most popular and accessible 3D representation, i.e., point clouds, involves discrete samples of the underlying continuous 3D surface. This discretization process introduces sampling variations on the 3D shape, making it challenging to develop transferable knowledge of the true 3D geometry. In the standard autoencoding paradigm, the encoder is compelled to encode not only the 3D geometry but also information on the specific discrete sampling of the 3D shape into the latent code. This is because the point cloud reconstructed by the decoder is considered unacceptable unless there is a perfect mapping between the original and the reconstructed point clouds. This paper introduces the Implicit AutoEncoder (IAE), a simple yet effective method that addresses the sampling variation issue by replacing the commonly-used point-cloud decoder with an implicit decoder. The implicit decoder reconstructs a continuous representation of the 3D shape, independent of the imperfections in the discrete samples. Extensive experiments demonstrate that the proposed IAE achieves state-of-the-art performance across various self-supervised learning benchmarks.

Yanjie Zhu, Haoxiang Li, Yuanyuan Liu et al.

Purpose: To propose a novel deep learning-based method called RG-Net (reconstruction and generation network) for highly accelerated MR parametric mapping by undersampling k-space and reducing the acquired contrast number simultaneously. Methods: The proposed framework consists of a reconstruction module and a generative module. The reconstruction module reconstructs MR images from the acquired few undersampled k-space data with the help of a data prior. The generative module then synthesizes the remaining multi-contrast images from the reconstructed images, where the exponential model is implicitly incorporated into the image generation through the supervision of fully sampled labels. The RG-Net was evaluated on the T1\r{ho} mapping data of knee and brain at different acceleration rates. Regional T1\r{ho} analysis for cartilage and the brain was performed to access the performance of RG-Net. Results: RG-Net yields a high-quality T1\r{ho} map at a high acceleration rate of 17. Compared with the competing methods that only undersample k-space, our framework achieves better performance in T1\r{ho} value analysis. Our method also improves quality of T1\r{ho} maps on patient with glioma. Conclusion: The proposed RG-Net that adopted a new strategy by undersampling k-space and reducing the contrast number simultaneously for fast MR parametric mapping, can achieve a high acceleration rate while maintaining good reconstruction quality. The generative module of our framework can also be used as an insert module in other fast MR parametric mapping methods. Keywords: Deep learning, convolutional neural network, fast MR parametric mapping

Yiding Yang, Zhou Ren, Haoxiang Li et al.

Multi-person pose estimation and tracking serve as crucial steps for video understanding. Most state-of-the-art approaches rely on first estimating poses in each frame and only then implementing data association and refinement. Despite the promising results achieved, such a strategy is inevitably prone to missed detections especially in heavily-cluttered scenes, since this tracking-by-detection paradigm is, by nature, largely dependent on visual evidences that are absent in the case of occlusion. In this paper, we propose a novel online approach to learning the pose dynamics, which are independent of pose detections in current fame, and hence may serve as a robust estimation even in challenging scenarios including occlusion. Specifically, we derive this prediction of dynamics through a graph neural network~(GNN) that explicitly accounts for both spatial-temporal and visual information. It takes as input the historical pose tracklets and directly predicts the corresponding poses in the following frame for each tracklet. The predicted poses will then be aggregated with the detected poses, if any, at the same frame so as to produce the final pose, potentially recovering the occluded joints missed by the estimator. Experiments on PoseTrack 2017 and PoseTrack 2018 datasets demonstrate that the proposed method achieves results superior to the state of the art on both human pose estimation and tracking tasks.

Bo Liu, Haoxiang Li, Hao Kang et al.

Main challenges in long-tailed recognition come from the imbalanced data distribution and sample scarcity in its tail classes. While techniques have been proposed to achieve a more balanced training loss and to improve tail classes data variations with synthesized samples, we resort to leverage readily available unlabeled data to boost recognition accuracy. The idea leads to a new recognition setting, namely semi-supervised long-tailed recognition. We argue this setting better resembles the real-world data collection and annotation process and hence can help close the gap to real-world scenarios. To address the semi-supervised long-tailed recognition problem, we present an alternate sampling framework combining the intuitions from successful methods in these two research areas. The classifier and feature embedding are learned separately and updated iteratively. The class-balanced sampling strategy has been implemented to train the classifier in a way not affected by the pseudo labels' quality on the unlabeled data. A consistency loss has been introduced to limit the impact from unlabeled data while leveraging them to update the feature embedding. We demonstrate significant accuracy improvements over other competitive methods on two datasets.

Bo Liu, Haoxiang Li, Hao Kang et al.

The problem of long-tailed recognition, where the number of examples per class is highly unbalanced, is considered. It is hypothesized that the well known tendency of standard classifier training to overfit to popular classes can be exploited for effective transfer learning. Rather than eliminating this overfitting, e.g. by adopting popular class-balanced sampling methods, the learning algorithm should instead leverage this overfitting to transfer geometric information from popular to low-shot classes. A new classifier architecture, GistNet, is proposed to support this goal, using constellations of classifier parameters to encode the class geometry. A new learning algorithm is then proposed for GeometrIc Structure Transfer (GIST), with resort to a combination of loss functions that combine class-balanced and random sampling to guarantee that, while overfitting to the popular classes is restricted to geometric parameters, it is leveraged to transfer class geometry from popular to few-shot classes. This enables better generalization for few-shot classes without the need for the manual specification of class weights, or even the explicit grouping of classes into different types. Experiments on two popular long-tailed recognition datasets show that GistNet outperforms existing solutions to this problem.

Bo Liu, Haoxiang Li, Hao Kang et al.

The problem of long-tailed recognition, where the number of examples per class is highly unbalanced, is considered. While training with class-balanced sampling has been shown effective for this problem, it is known to over-fit to few-shot classes. It is hypothesized that this is due to the repeated sampling of examples and can be addressed by feature space augmentation. A new feature augmentation strategy, EMANATE, based on back-tracking of features across epochs during training, is proposed. It is shown that, unlike class-balanced sampling, this is an adversarial augmentation strategy. A new sampling procedure, Breadcrumb, is then introduced to implement adversarial class-balanced sampling without extra computation. Experiments on three popular long-tailed recognition datasets show that Breadcrumb training produces classifiers that outperform existing solutions to the problem.

Wei Wei, Li Guan, Yue Liu et al.

HDR reconstruction is an important task in computer vision with many industrial needs. The traditional approaches merge multiple exposure shots to generate HDRs that correspond to the physical quantity of illuminance of the scene. However, the tedious capturing process makes such multi-shot approaches inconvenient in practice. In contrast, recent single-shot methods predict a visually appealing HDR from a single LDR image through deep learning. But it is not clear whether the previously mentioned physical properties would still hold, without training the network to explicitly model them. In this paper, we introduce the physical illuminance constraints to our single-shot HDR reconstruction framework, with a focus on spherical panoramas. By the proposed physical regularization, our method can generate HDRs which are not only visually appealing but also physically plausible. For evaluation, we collect a large dataset of LDR and HDR images with ground truth illuminance measures. Extensive experiments show that our HDR images not only maintain high visual quality but also top all baseline methods in illuminance prediction accuracy.

Bo Liu, Hao Kang, Haoxiang Li et al.

The problem of open-set recognition is considered. While previous approaches only consider this problem in the context of large-scale classifier training, we seek a unified solution for this and the low-shot classification setting. It is argued that the classic softmax classifier is a poor solution for open-set recognition, since it tends to overfit on the training classes. Randomization is then proposed as a solution to this problem. This suggests the use of meta-learning techniques, commonly used for few-shot classification, for the solution of open-set recognition. A new oPen sEt mEta LEaRning (PEELER) algorithm is then introduced. This combines the random selection of a set of novel classes per episode, a loss that maximizes the posterior entropy for examples of those classes, and a new metric learning formulation based on the Mahalanobis distance. Experimental results show that PEELER achieves state of the art open set recognition performance for both few-shot and large-scale recognition. On CIFAR and miniImageNet, it achieves substantial gains in seen/unseen class detection AUROC for a given seen-class classification accuracy.

Hao Kang, Jianming Zhang, Haoxiang Li et al.

We introduce a novel framework for automatic capturing of human portraits. The framework allows the robot to follow a person to the desired location using a Person Re-identification model. When composing is activated, the robot attempts to adjust its position to form the view that can best match the given template image, and finally takes a photograph. A template image can be predicted dynamically using an off-the-shelf photo evaluation model by the framework, or selected manually from a pre-defined set by the user. The template matching-based view adjustment is driven by a deep reinforcement learning network. Our framework lies on top of the Robot Operating System (ROS). The framework is designed to be modular so that all the models can be flexibly replaced based on needs. We show our framework on a variety of examples. In particular, we tested it in three indoor scenes and used it to take 20 photos of each scene: ten for the pre-defined template, ten for the dynamically generated ones. The average number of adjustment was $11.20$ for pre-defined templates and $12.76$ for dynamically generated ones; the average time spent was $22.11$ and $24.10$ seconds respectively.

Yinan Zhang, Devin Balkcom, Haoxiang Li

This paper addresses the question of how a previously available control policy $π_s$ can be used as a supervisor to more quickly and safely train a new learned control policy $π_L$ for a robot. A weighted average of the supervisor and learned policies is used during trials, with a heavier weight initially on the supervisor, in order to allow safe and useful physical trials while the learned policy is still ineffective. During the process, the weight is adjusted to favor the learned policy. As weights are adjusted, the learned network must compensate so as to give safe and reasonable outputs under the different weights. A pioneer network is introduced that pre-learns a policy that performs similarly to the current learned policy under the planned next step for new weights; this pioneer network then replaces the currently learned network in the next set of trials. Experiments in OpenAI Gym demonstrate the effectiveness of the proposed method.

Rameswar Panda, Jianming Zhang, Haoxiang Li et al.

While machine learning approaches to visual emotion recognition offer great promise, current methods consider training and testing models on small scale datasets covering limited visual emotion concepts. Our analysis identifies an important but long overlooked issue of existing visual emotion benchmarks in the form of dataset biases. We design a series of tests to show and measure how such dataset biases obstruct learning a generalizable emotion recognition model. Based on our analysis, we propose a webly supervised approach by leveraging a large quantity of stock image data. Our approach uses a simple yet effective curriculum guided training strategy for learning discriminative emotion features. We discover that the models learned using our large scale stock image dataset exhibit significantly better generalization ability than the existing datasets without the manual collection of even a single label. Moreover, visual representation learned using our approach holds a lot of promise across a variety of tasks on different image and video datasets.

Xin Ye, Zhe Lin, Haoxiang Li et al.

We study the problem of learning a navigation policy for a robot to actively search for an object of interest in an indoor environment solely from its visual inputs. While scene-driven visual navigation has been widely studied, prior efforts on learning navigation policies for robots to find objects are limited. The problem is often more challenging than target scene finding as the target objects can be very small in the view and can be in an arbitrary pose. We approach the problem from an active perceiver perspective, and propose a novel framework that integrates a deep neural network based object recognition module and a deep reinforcement learning based action prediction mechanism. To validate our method, we conduct experiments on both a simulation dataset (AI2-THOR) and a real-world environment with a physical robot. We further propose a new decaying reward function to learn the control policy specific to the object searching task. Experimental results validate the efficacy of our method, which outperforms competing methods in both average trajectory length and success rate.

Xiangyun Zhao, Haoxiang Li, Xiaohui Shen et al.

Multi-task learning has been widely adopted in many computer vision tasks to improve overall computation efficiency or boost the performance of individual tasks, under the assumption that those tasks are correlated and complementary to each other. However, the relationships between the tasks are complicated in practice, especially when the number of involved tasks scales up. When two tasks are of weak relevance, they may compete or even distract each other during joint training of shared parameters, and as a consequence undermine the learning of all the tasks. This will raise destructive interference which decreases learning efficiency of shared parameters and lead to low quality loss local optimum w.r.t. shared parameters. To address the this problem, we propose a general modulation module, which can be inserted into any convolutional neural network architecture, to encourage the coupling and feature sharing of relevant tasks while disentangling the learning of irrelevant tasks with minor parameters addition. Equipped with this module, gradient directions from different tasks can be enforced to be consistent for those shared parameters, which benefits multi-task joint training. The module is end-to-end learnable without ad-hoc design for specific tasks, and can naturally handle many tasks at the same time. We apply our approach on two retrieval tasks, face retrieval on the CelebA dataset [1] and product retrieval on the UT-Zappos50K dataset [2, 3], and demonstrate its advantage over other multi-task learning methods in both accuracy and storage efficiency.

Bangjie Yin, Luan Tran, Haoxiang Li et al.

Deep CNNs have been pushing the frontier of visual recognition over past years. Besides recognition accuracy, strong demands in understanding deep CNNs in the research community motivate developments of tools to dissect pre-trained models to visualize how they make predictions. Recent works further push the interpretability in the network learning stage to learn more meaningful representations. In this work, focusing on a specific area of visual recognition, we report our efforts towards interpretable face recognition. We propose a spatial activation diversity loss to learn more structured face representations. By leveraging the structure, we further design a feature activation diversity loss to push the interpretable representations to be discriminative and robust to occlusions. We demonstrate on three face recognition benchmarks that our proposed method is able to improve face recognition accuracy with easily interpretable face representations.

Kaichun Mo, Haoxiang Li, Zhe Lin et al.

Deep reinforcement learning (DRL) demonstrates its potential in learning a model-free navigation policy for robot visual navigation. However, the data-demanding algorithm relies on a large number of navigation trajectories in training. Existing datasets supporting training such robot navigation algorithms consist of either 3D synthetic scenes or reconstructed scenes. Synthetic data suffers from domain gap to the real-world scenes while visual inputs rendered from 3D reconstructed scenes have undesired holes and artifacts. In this paper, we present a new dataset collected in real-world to facilitate the research in DRL based visual navigation. Our dataset includes 3D reconstruction for real-world scenes as well as densely captured real 2D images from the scenes. It provides high-quality visual inputs with real-world scene complexity to the robot at dense grid locations. We further study and benchmark one recent DRL based navigation algorithm and present our attempts and thoughts on improving its generalizability to unseen test targets in the scenes.

Ronald Yu, Shunsuke Saito, Haoxiang Li et al.

We present a minimalistic but effective neural network that computes dense facial correspondences in highly unconstrained RGB images. Our network learns a per-pixel flow and a matchability mask between 2D input photographs of a person and the projection of a textured 3D face model. To train such a network, we generate a massive dataset of synthetic faces with dense labels using renderings of a morphable face model with variations in pose, expressions, lighting, and occlusions. We found that a training refinement using real photographs is required to drastically improve the ability to handle real images. When combined with a facial detection and 3D face fitting step, we show that our approach outperforms the state-of-the-art face alignment methods in terms of accuracy and speed. By directly estimating dense correspondences, we do not rely on the full visibility of sparse facial landmarks and are not limited to the model space of regression-based approaches. We also assess our method on video frames and demonstrate successful per-frame processing under extreme pose variations, occlusions, and lighting conditions. Compared to existing 3D facial tracking techniques, our fitting does not rely on previous frames or frontal facial initialization and is robust to imperfect face detections.

Chuang Gan, Yandong Li, Haoxiang Li et al.

Rich and dense human labeled datasets are among the main enabling factors for the recent advance on vision-language understanding. Many seemingly distant annotations (e.g., semantic segmentation and visual question answering (VQA)) are inherently connected in that they reveal different levels and perspectives of human understandings about the same visual scenes --- and even the same set of images (e.g., of COCO). The popularity of COCO correlates those annotations and tasks. Explicitly linking them up may significantly benefit both individual tasks and the unified vision and language modeling. We present the preliminary work of linking the instance segmentations provided by COCO to the questions and answers (QAs) in the VQA dataset, and name the collected links visual questions and segmentation answers (VQS). They transfer human supervision between the previously separate tasks, offer more effective leverage to existing problems, and also open the door for new research problems and models. We study two applications of the VQS data in this paper: supervised attention for VQA and a novel question-focused semantic segmentation task. For the former, we obtain state-of-the-art results on the VQA real multiple-choice task by simply augmenting the multilayer perceptrons with some attention features that are learned using the segmentation-QA links as explicit supervision. To put the latter in perspective, we study two plausible methods and compare them to an oracle method assuming that the instance segmentations are given at the test stage.