Jikai Wang

Shuwei Li, Jikai Wang, Michael S. Brown et al.

The assumption of a uniform light color distribution is no longer applicable in scenes that have multiple light colors. Most color constancy methods are designed to deal with a single light color, and thus are erroneous when applied to multiple light colors. The spatial variability in multiple light colors causes the color constancy problem to be more challenging and requires the extraction of local surface/light information. Motivated by this, we introduce a multi-task learning method to discount multiple light colors in a single input image. To have better cues of the local surface/light colors under multiple light color conditions, we design a novel multi-task learning framework. Our framework includes auxiliary tasks of achromatic-pixel detection and surface-color similarity prediction, providing better cues for local light and surface colors, respectively. Moreover, to ensure that our model maintains the constancy of surface colors regardless of the variations of light colors, a novel local surface color feature preservation scheme is developed. We demonstrate that our model achieves 47.1% improvement (from 4.69 mean angular error to 2.48) compared to a state-of-the-art multi-illuminant color constancy method on a multi-illuminant dataset (LSMI).

Alakh Aggarwal, Jikai Wang, Steven Hogue et al.

Recent research works have focused on generating human models and garments from their 2D images. However, state-of-the-art researches focus either on only a single layer of the garment on a human model or on generating multiple garment layers without any guarantee of the intersection-free geometric relationship between them. In reality, people wear multiple layers of garments in their daily life, where an inner layer of garment could be partially covered by an outer one. In this paper, we try to address this multi-layer modeling problem and propose the Layered-Garment Net (LGN) that is capable of generating intersection-free multiple layers of garments defined by implicit function fields over the body surface, given the person's near front-view image. With a special design of garment indication fields (GIF), we can enforce an implicit covering relationship between the signed distance fields (SDF) of different layers to avoid self-intersections among different garment surfaces and the human body. Experiments demonstrate the strength of our proposed LGN framework in generating multi-layer garments as compared to state-of-the-art methods. To the best of our knowledge, LGN is the first research work to generate intersection-free multiple layers of garments on the human body from a single image.

Wanglong Lu, Jikai Wang, Tao Wang et al.

Blind face restoration aims to recover high-quality facial images from various unidentified sources of degradation, posing significant challenges due to the minimal information retrievable from the degraded images. Prior knowledge-based methods, leveraging geometric priors and facial features, have led to advancements in face restoration but often fall short of capturing fine details. To address this, we introduce a visual style prompt learning framework that utilizes diffusion probabilistic models to explicitly generate visual prompts within the latent space of pre-trained generative models. These prompts are designed to guide the restoration process. To fully utilize the visual prompts and enhance the extraction of informative and rich patterns, we introduce a style-modulated aggregation transformation layer. Extensive experiments and applications demonstrate the superiority of our method in achieving high-quality blind face restoration. The source code is available at \href{https://github.com/LonglongaaaGo/VSPBFR}{https://github.com/LonglongaaaGo/VSPBFR}.

Jikai Wang, Juntao Li, Jianye Hou et al.

Large reasoning language models such as OpenAI-o1 and Deepseek-R1 have recently attracted widespread attention due to their impressive task-solving abilities. However, the enormous model size and the generation of lengthy thought chains introduce significant reasoning costs and response latency. Existing methods for efficient reasoning mainly focus on reducing the number of model parameters or shortening the chain-of-thought length. In this paper, we introduce Speculative Chain-of-Thought (SCoT), which reduces reasoning latency from another perspective by accelerated average reasoning speed through large and small model collaboration. SCoT conducts thought-level drafting using a lightweight draft model. Then it selects the best CoT draft and corrects the error cases with the target model. The proposed thinking behavior alignment improves the efficiency of drafting and the draft selection strategy maintains the prediction accuracy of the target model for complex tasks. Experimental results on GSM8K, MATH, GaoKao, CollegeMath and Olympiad datasets show that SCoT reduces reasoning latency by 48\%$\sim$66\% and 21\%$\sim$49\% for Deepseek-R1-Distill-Qwen-32B and Deepseek-R1-Distill-Llama-70B while achieving near-target-model-level performance. Our code is available at https://github.com/Jikai0Wang/Speculative_CoT.

Jikai Wang, Jianchao Tan, Yuxuan Hu et al.

Speculative decoding accelerates autoregressive language model inference by verifying multiple draft tokens in parallel. However, the verification stage often becomes the dominant computational bottleneck, especially for long-context inputs and mixture-of-experts (MoE) models. Existing sparsification methods are designed primarily for standard token-by-token autoregressive decoding to remove substantial computational redundancy in LLMs. This work systematically adopts different sparse methods on the verification stage of the speculative decoding and identifies structured redundancy across multiple dimensions. Based on these observations, we propose a sparse verification framework that jointly sparsifies attention, FFN, and MoE components during the verification stage to reduce the dominant computation cost. The framework further incorporates an inter-draft token and inter-layer retrieval reuse strategy to further reduce redundant computation without introducing additional training. Extensive experiments across summarization, question answering, and mathematical reasoning datasets demonstrate that the proposed methods achieve favorable efficiency-accuracy trade-offs, while maintaining stable acceptance length.

Xingtai Gui, Jianbo Zhao, Wencheng Han et al.

End-to-end autonomous driving systems directly generate driving policies from raw sensor inputs. While these systems can extract effective environmental features for planning, relying on auxiliary perception tasks, developing perception annotation-free planning paradigms has become increasingly critical due to the high cost of manual perception annotation. In this work, we propose TrajDiff, a Trajectory-oriented BEV Conditioned Diffusion framework that establishes a fully perception annotation-free generative method for end-to-end autonomous driving. TrajDiff requires only raw sensor inputs and future trajectory, constructing Gaussian BEV heatmap targets that inherently capture driving modalities. We design a simple yet effective trajectory-oriented BEV encoder to extract the TrajBEV feature without perceptual supervision. Furthermore, we introduce Trajectory-oriented BEV Diffusion Transformer (TB-DiT), which leverages ego-state information and the predicted TrajBEV features to directly generate diverse yet plausible trajectories, eliminating the need for handcrafted motion priors. Beyond architectural innovations, TrajDiff enables exploration of data scaling benefits in the annotation-free setting. Evaluated on the NAVSIM benchmark, TrajDiff achieves 87.5 PDMS, establishing state-of-the-art performance among all annotation-free methods. With data scaling, it further improves to 88.5 PDMS, which is comparable to advanced perception-based approaches. Our code and model will be made publicly available.

Jishnu Jaykumar P, Cole Salvato, Vinaya Bomnale et al.

Robotic perception models often fail when deployed in real-world environments due to out-of-distribution conditions such as clutter, occlusion, and novel object instances. Existing approaches address this gap through offline data collection and retraining, which are slow and do not resolve deployment-time failures. We propose iTeach, a failure-driven interactive teaching framework for adapting robot perception in the wild. A co-located human observes model predictions during deployment, identifies failure cases, and performs short human-object interaction (HumanPlay) to expose informative object configurations while recording RGB-D sequences. To minimize annotation effort, iTeach employs a Few-Shot Semi- Supervised (FS3) labeling strategy, where only the final frame of a short interaction sequence is annotated using hands-free eye-gaze and voice commands, and labels are propagated across the video to produce dense supervision. The collected failure-driven samples are used for iterative fine-tuning, enabling progressive deployment-time adaptation of the perception model. We evaluate iTeach on unseen object instance segmentation (UOIS) starting from a pretrained MSMFormer model. Using a small number of failure-driven samples, our method significantly improves segmentation performance across diverse real-world scenes. These improvements directly translate to higher grasping and pick-and-place success on the SceneReplica benchmark and real robotic experiments. Our results demonstrate that failure-driven, co-located interactive teaching enables efficient in-the-wild adaptation of robot perception and improves downstream manipulation performance. Project page at https://irvlutd.github.io/iTeach

Qifan Zhang, Sai Haneesh Allu, Jikai Wang et al.

Detecting and segmenting novel object instances in open-world environments is a fundamental problem in robotic perception. Given only a small set of template images, a robot must locate and segment a specific object instance in a cluttered, previously unseen scene. Existing proposal-based approaches are highly sensitive to proposal quality and often fail under occlusion and background clutter. We propose L2G-Det, a local-to-global instance detection framework that bypasses explicit object proposals by leveraging dense patch-level matching between templates and the query image. Locally matched patches generate candidate points, which are refined through a candidate selection module to suppress false positives. The filtered points are then used to prompt an augmented Segment Anything Model (SAM) with instance-specific object tokens, enabling reliable reconstruction of complete instance masks. Experiments demonstrate improved performance over proposal-based methods in challenging open-world settings.

Jikai Wang, Yi Su, Juntao Li et al.

Autoregressive language models demonstrate excellent performance in various scenarios. However, the inference efficiency is limited by its one-step-one-word generation mode, which has become a pressing problem recently as the models become increasingly larger. Speculative decoding employs a "draft and then verify" mechanism to allow multiple tokens to be generated in one step, realizing lossless acceleration. Existing methods mainly adopt fixed heuristic draft structures, which fail to adapt to different situations to maximize the acceptance length during verification. To alleviate this dilemma, we proposed OPT-Tree, an algorithm to construct adaptive and scalable draft trees. It searches the optimal tree structure that maximizes the mathematical expectation of the acceptance length in each decoding step. Experimental results reveal that OPT-Tree outperforms the existing draft structures and achieves a speed-up ratio of up to 3.2 compared with autoregressive decoding. If the draft model is powerful enough and the node budget is sufficient, it can generate more than ten tokens in a single step. Our code is available at https://github.com/Jikai0Wang/OPT-Tree.

Dan Qiao, Yi Su, Pinzheng Wang et al.

Large Language Models (LLMs) have played an important role in many fields due to their powerful capabilities.However, their massive number of parameters leads to high deployment requirements and incurs significant inference costs, which impedes their practical applications. Training smaller models is an effective way to address this problem. Therefore, we introduce OpenBA-V2, a 3.4B model derived from multi-stage compression and continual pre-training from the original 15B OpenBA model. OpenBA-V2 utilizes more data, more flexible training objectives, and techniques such as layer pruning, neural pruning, and vocabulary pruning to achieve a compression rate of 77.3\% with minimal performance loss. OpenBA-V2 demonstrates competitive performance compared to other open-source models of similar size, achieving results close to or on par with the 15B OpenBA model in downstream tasks such as common sense reasoning and Named Entity Recognition (NER). OpenBA-V2 illustrates that LLMs can be compressed into smaller ones with minimal performance loss by employing advanced training objectives and data strategies, which may help deploy LLMs in resource-limited scenarios.

Rohith Peddi, Shivvrat Arya, Bharath Challa et al.

Following step-by-step procedures is an essential component of various activities carried out by individuals in their daily lives. These procedures serve as a guiding framework that helps to achieve goals efficiently, whether it is assembling furniture or preparing a recipe. However, the complexity and duration of procedural activities inherently increase the likelihood of making errors. Understanding such procedural activities from a sequence of frames is a challenging task that demands an accurate interpretation of visual information and the ability to reason about the structure of the activity. To this end, we collect a new egocentric 4D dataset, CaptainCook4D, comprising 384 recordings (94.5 hours) of people performing recipes in real kitchen environments. This dataset consists of two distinct types of activity: one in which participants adhere to the provided recipe instructions and another in which they deviate and induce errors. We provide 5.3K step annotations and 10K fine-grained action annotations and benchmark the dataset for the following tasks: supervised error recognition, multistep localization, and procedure learning

Wanglong Lu, Jikai Wang, Xiaogang Jin et al.

Existing facial editing methods have achieved remarkable results, yet they often fall short in supporting multimodal conditional local facial editing. One of the significant evidences is that their output image quality degrades dramatically after several iterations of incremental editing, as they do not support local editing. In this paper, we present a novel multimodal generative and fusion framework for globally-consistent local facial editing (FACEMUG) that can handle a wide range of input modalities and enable fine-grained and semantic manipulation while remaining unedited parts unchanged. Different modalities, including sketches, semantic maps, color maps, exemplar images, text, and attribute labels, are adept at conveying diverse conditioning details, and their combined synergy can provide more explicit guidance for the editing process. We thus integrate all modalities into a unified generative latent space to enable multimodal local facial edits. Specifically, a novel multimodal feature fusion mechanism is proposed by utilizing multimodal aggregation and style fusion blocks to fuse facial priors and multimodalities in both latent and feature spaces. We further introduce a novel self-supervised latent warping algorithm to rectify misaligned facial features, efficiently transferring the pose of the edited image to the given latent codes. We evaluate our FACEMUG through extensive experiments and comparisons to state-of-the-art (SOTA) methods. The results demonstrate the superiority of FACEMUG in terms of editing quality, flexibility, and semantic control, making it a promising solution for a wide range of local facial editing tasks.

Shibowen Zhang, Jiayang Wu, Guannan Liu et al.

This paper presents an integrated model-based framework for generating and executing dynamic whole-body dance motions on humanoid robots. The framework operates in two stages: offline motion generation and online motion execution, both leveraging future state prediction to enable robust and dynamic dance motions in real-world environments. In the offline motion generation stage, human dance demonstrations are captured via a motion capture (MoCap) system, retargeted to the robot by solving a Quadratic Programming (QP) problem, and further refined using Trajectory Optimization (TO) to ensure dynamic feasibility. In the online motion execution stage, a centroidal dynamics-based Model Predictive Control (MPC) framework tracks the planned motions in real time and proactively adjusts swing foot placement to adapt to real world disturbances. We validate our framework on the full-size humanoid robot Kuavo 4Pro, demonstrating the dynamic dance motions both in simulation and in a four-minute live public performance with a team of four robots. Experimental results show that longer prediction horizons improve both motion expressiveness in planning and stability in execution.

Jikai Wang, Zhenxu Tian, Juntao Li et al.

Recent works have revealed the great potential of speculative decoding in accelerating the autoregressive generation process of large language models. The success of these methods relies on the alignment between draft candidates and the sampled outputs of the target model. Existing methods mainly achieve draft-target alignment with training-based methods, e.g., EAGLE, Medusa, involving considerable training costs. In this paper, we present a training-free alignment-augmented speculative decoding algorithm. We propose alignment sampling, which leverages output distribution obtained in the prefilling phase to provide more aligned draft candidates. To further benefit from high-quality but non-aligned draft candidates, we also introduce a simple yet effective flexible verification strategy. Through an adaptive probability threshold, our approach can improve generation accuracy while further improving inference efficiency. Experiments on 8 datasets (including question answering, summarization and code completion tasks) show that our approach increases the average generation score by 3.3 points for the LLaMA3 model. Our method achieves a mean acceptance length up to 2.39 and speed up generation by 2.23.

Yangxiao Lu, Ruosen Li, Liqiang Jing et al.

Visual grounding focuses on detecting objects from images based on language expressions. Recent Large Vision-Language Models (LVLMs) have significantly advanced visual grounding performance by training large models with large-scale datasets. However, the problem remains challenging, especially when similar objects appear in the input image. For example, an LVLM may not be able to differentiate Diet Coke and regular Coke in an image. In this case, if additional reference images of Diet Coke and regular Coke are available, it can help the visual grounding of similar objects. In this work, we introduce a new task named Multimodal Reference Visual Grounding (MRVG). In this task, a model has access to a set of reference images of objects in a database. Based on these reference images and a language expression, the model is required to detect a target object from a query image. We first introduce a new dataset to study the MRVG problem. Then we introduce a novel method, named MRVG-Net, to solve this visual grounding problem. We show that by efficiently using reference images with few-shot object detection and using Large Language Models (LLMs) for object matching, our method achieves superior visual grounding performance compared to the state-of-the-art LVLMs such as Qwen2.5-VL-72B. Our approach bridges the gap between few-shot detection and visual grounding, unlocking new capabilities for visual understanding, which has wide applications in robotics. Project page with our video, code, and dataset: https://irvlutd.github.io/MultiGrounding

Jikai Wang, Huan Zheng, Jianbing Shen

Lightweight image super-resolution (SR) methods aim at increasing the resolution and restoring the details of an image using a lightweight neural network. However, current lightweight SR methods still suffer from inferior performance and unpleasant details. Our analysis reveals that these methods are hindered by constrained feature diversity, which adversely impacts feature representation and detail recovery. To respond this issue, we propose a simple yet effective baseline called CubeFormer, designed to enhance feature richness by completing holistic information aggregation. To be specific, we introduce cube attention, which expands 2D attention to 3D space, facilitating exhaustive information interactions, further encouraging comprehensive information extraction and promoting feature variety. In addition, we inject block and grid sampling strategies to construct intra-cube transformer blocks (Intra-CTB) and inter-cube transformer blocks (Inter-CTB), which perform local and global modeling, respectively. Extensive experiments show that our CubeFormer achieves state-of-the-art performance on commonly used SR benchmarks. Our source code and models will be publicly available.

Jikai Wang, Qifan Zhang, Yu-Wei Chao et al.

We introduce a data capture system and a new dataset, HO-Cap, for 3D reconstruction and pose tracking of hands and objects in videos. The system leverages multiple RGBD cameras and a HoloLens headset for data collection, avoiding the use of expensive 3D scanners or mocap systems. We propose a semi-automatic method for annotating the shape and pose of hands and objects in the collected videos, significantly reducing the annotation time compared to manual labeling. With this system, we captured a video dataset of humans interacting with objects to perform various tasks, including simple pick-and-place actions, handovers between hands, and using objects according to their affordance, which can serve as human demonstrations for research in embodied AI and robot manipulation. Our data capture setup and annotation framework will be available for the community to use in reconstructing 3D shapes of objects and human hands and tracking their poses in videos.

Peng Bao, Zonghai Chen, Jikai Wang et al.

Accurate trajectory prediction of vehicles is essential for reliable autonomous driving. To maintain consistent performance as a vehicle driving around different cities, it is crucial to adapt to changing traffic circumstances and achieve lifelong trajectory prediction model. To realize it, catastrophic forgetting is a main problem to be addressed. In this paper, a divergence measurement method based on conditional Kullback-Leibler divergence is proposed first to evaluate spatiotemporal dependency difference among varied driving circumstances. Then based on generative replay, a novel lifelong vehicle trajectory prediction framework is developed. The framework consists of a conditional generation model and a vehicle trajectory prediction model. The conditional generation model is a generative adversarial network conditioned on position configuration of vehicles. After learning and merging trajectory distribution of vehicles across different cities, the generation model replays trajectories with prior samplings as inputs, which alleviates catastrophic forgetting. The vehicle trajectory prediction model is trained by the replayed trajectories and achieves consistent prediction performance on visited cities. A lifelong experiment setup is established on four open datasets including five tasks. Spatiotemporal dependency divergence is calculated for different tasks. Even though these divergence, the proposed framework exhibits lifelong learning ability and achieves consistent performance on all tasks.

Zhenkun Zhu, Jikai Wang

In recent years, visual SLAM has achieved great progress and development in different scenes, however, there are still many problems to be solved. The SLAM system is not only restricted by the external scenes but is also affected by its movement mode, such as movement speed, rotational motion, etc. As the representatives of the most excellent networks for frame interpolation, Sepconv-slomo and EDSC can predict high-quality intermediate frame between the previous frame and the current frame. Intuitively, frame interpolation technology can enrich the information of images sequences, the number of which is limited by the camera's frame rate, and thus decreasing the probability of SLAM system's failure rate. In this article, we propose an InterpolationSLAM framework. InterpolationSLAM is robust in rotational movement for Monocular and RGB-D configurations. By detecting the rotation and performing interpolation processing at the rotated position, pose of the system can be estimated more accurately, thereby improving the accuracy and robustness of the SLAM system in the rotational movement.

Zhenkun Zhu, Jikai Wang

In recent years, visual SLAM has achieved great progress and development, but in complex scenes, especially rotating scenes, the error of mapping will increase significantly, and the slam system is easy to lose track. In this article, we propose an InterpolationSLAM framework, which is a visual SLAM framework based on ORB-SLAM2. InterpolationSLAM is robust in rotating scenes for Monocular and RGB-D configurations. By detecting the rotation and performing interpolation processing at the rotated position, pose of the system can be estimated more accurately at the rotated position, thereby improving the accuracy and robustness of the SLAM system in the rotating scenes. To the best of our knowledge, it is the first work combining the interpolation network into a Visual SLAM system to improve SLAM system robustness in rotating scenes. We conduct experiments both on KITTI Monocular and TUM RGB-D datasets. The results demonstrate that InterpolationSLAM outperforms the accuracy of standard Visual SLAM baselines.