Yayuan Li

Yayuan Li, Jian Zhang, Jintao Guo et al.

The proliferation of pre-trained models has given rise to a wide array of specialised, fine-tuned models. Model merging aims to merge the distinct capabilities of these specialised models into a unified model, requiring minimal or even no additional training. A core objective of model merging is to ensure the merged model retains the behavioural characteristics of the specialised models, typically achieved through feature alignment. We identify that features consist of two critical components: direction and magnitude. Prior research has predominantly focused on directional alignment, while the influence of magnitude remains largely neglected, despite its pronounced vulnerability to perturbations introduced by common merging operations (e.g., parameter fusion and sparsification). Such perturbations to magnitude inevitably lead to feature deviations in the merged model from the specialised models, resulting in subsequent performance degradation. To address this, we propose MAGnItude Calibration (MAGIC), a plug-and-play framework that rectifies layer-wise magnitudes in feature and weight spaces, with three variants. Specifically, our Feature Space Calibration (FSC) realigns the merged model's features using a small set of unlabelled data, while Weight Space Calibration (WSC) extends this calibration to the weight space without requiring additional data. Combining these yields Dual Space Calibration (DSC). Comprehensive experiments demonstrate that MAGIC consistently boosts performance across diverse Computer Vision tasks (+4.3% on eight datasets) and NLP tasks (+8.0% on Llama) without additional training. Our code is available at: https://github.com/lyymuwu/MAGIC

Yayuan Li, Ze Peng, Jian Zhang et al.

Model merging combines multiple fine-tuned models into a single model by adding their weight updates, providing a lightweight alternative to retraining. Existing methods primarily target resolving conflicts between task updates, leaving the failure mode of over-counting shared knowledge unaddressed. We show that when tasks share aligned spectral directions (i.e., overlapping singular vectors), a simple linear combination repeatedly accumulates these directions, inflating the singular values and biasing the merged model toward shared subspaces. To mitigate this issue, we propose Singular Value Calibration (SVC), a training-free and data-free post-processing method that quantifies subspace overlap and rescales inflated singular values to restore a balanced spectrum. Across vision and language benchmarks, SVC consistently improves strong merging baselines and achieves state-of-the-art performance. Furthermore, by modifying only the singular values, SVC improves the performance of Task Arithmetic by 13.0%. Code is available at: https://github.com/lyymuwu/SVC.

Filippos Bellos, Yutong Li, Jessie N Dong et al.

Point-of-care transthoracic echocardiography (TTE) enables cardiac assessment in virtually any clinical setting, yet its diagnostic utility remains constrained by the expertise required for image acquisition and interpretation. Visual question answering (VQA) offers a promising paradigm for bridging this expertise gap through interactive clinical assistance, but existing echocardiography VQA datasets are limited in scale, restricted to high-quality images, and only cover a few views. We introduce EchoVQA, the first large-scale VQA dataset for echocardiography, comprising 14,299 images and 74,819 question-answer pairs. The dataset integrates public sources (EchoNet-Dynamic, CAMUS) with our own point-of-care acquisitions from two handheld probes (Lumify, Clarius), spanning diverse views and including both high-quality and suboptimal images. Uniquely, EchoVQA includes acquisition guidance questions to help users optimize transducer positioning toward a diagnostic apical 4-chamber view for left ventricular ejection fraction estimation -- a challenging task for novice operators in point-of-care settings. We further develop a parameter-efficient method based on multimodal learnable prompts achieving state-of-the-art performance on most benchmarks, including EchoVQA, with significantly less trainable parameters than existing state-of-the-art approaches.

Yayuan Li, Chenglin Li, Jingying Wang et al.

Instructional videos are the dominant medium for learning physical tasks, yet they rarely match the user's real-world visual context. Motor simulation and cognitive load theories predict this mismatch should matter, but we do not know (1) how much it could affect task completion, (2) which visual attributes are responsible, and (3) how users experience it. We conduct two complementary studies (56 participants, 86+ hours, four first-aid and culinary tasks) in which we use Wizard-of-Oz recordings to control the degree of visual alignment in instructional videos. In Study 1 (N=16), we prepare In-Context instructional videos (ICON) -- fully aligned with the user's visual perception -- to compare against business-as-usual Internet videos. ICON yields statistically significant improvements: 11.1% higher completion quality and 15.5% faster completion. Qualitative analysis reveals four visual context attributes responsible for the effect: Task Object Intrinsics, Task Object State, Environmental Context, and Observational Context. Study 2 (N=40) ablates each attribute by systematically misaligning one at a time from an otherwise fully aligned video, confirming all four produce consistent degradation. However, we find users fail to perceive the effect of single-attribute misalignment on task performance despite clear drops in objective measurement. Visual context misalignment is substantial, decomposable, and invisible to the user. These findings help understand the effect of visual context mismatch and how we should evaluate instructional videos for physical task guidance.

Yayuan Li, Jintao Guo, Lei Qi et al.

Contrastive Language-Image Pretraining (CLIP) has been widely used in vision tasks. Notably, CLIP has demonstrated promising performance in few-shot learning (FSL). However, existing CLIP-based methods in training-free FSL (i.e., without the requirement of additional training) mainly learn different modalities independently, leading to two essential issues: 1) severe anomalous match in image modality; 2) varying quality of generated text prompts. To address these issues, we build a mutual guidance mechanism, that introduces an Image-Guided-Text (IGT) component to rectify varying quality of text prompts through image representations, and a Text-Guided-Image (TGI) component to mitigate the anomalous match of image modality through text representations. By integrating IGT and TGI, we adopt a perspective of Text-Image Mutual guidance Optimization, proposing TIMO. Extensive experiments show that TIMO significantly outperforms the state-of-the-art (SOTA) training-free method. Additionally, by exploring the extent of mutual guidance, we propose an enhanced variant, TIMO-S, which even surpasses the best training-required methods by 0.33% with approximately 100 times less time cost. Our code is available at https://github.com/lyymuwu/TIMO.

Filippos Bellos, Yayuan Li, Cary Shu et al.

Effective human-AI collaboration for physical task completion has significant potential in both everyday activities and professional domains. AI agents equipped with informative guidance can enhance human performance, but evaluating such collaboration remains challenging due to the complexity of human-in-the-loop interactions. In this work, we introduce an evaluation framework and a multimodal dataset of human-AI interactions designed to assess how AI guidance affects procedural task performance, error reduction and learning outcomes. Besides, we develop an augmented reality (AR)-equipped AI agent that provides interactive guidance in real-world tasks, from cooking to battlefield medicine. Through human studies, we share empirical insights into AI-assisted human performance and demonstrate that AI-assisted collaboration improves task completion.

Shane Storks, Itamar Bar-Yossef, Yayuan Li et al.

Procedural mistake detection (PMD) is a challenging problem of classifying whether a human user (observed through egocentric video) has successfully executed a task (specified by a procedural text). Despite significant recent efforts, machine performance in the wild remains nonviable, and the reasoning processes underlying this performance are opaque. As such, we extend PMD to require generating visual self-dialog rationales to inform decisions. Given the impressive, mature image understanding capabilities observed in recent vision-and-language models (VLMs), we curate a suitable benchmark dataset for PMD based on individual frames. As our reformulation enables unprecedented transparency, we leverage a natural language inference (NLI) model to formulate two automated metrics for the coherence of generated rationales. We establish baselines for this reframed task, showing that VLMs struggle off-the-shelf, but with some trade-offs, their accuracy, coherence, and efficiency can be improved by incorporating these metrics into common inference and fine-tuning methods. Lastly, our multi-faceted metrics visualize common outcomes, highlighting areas for further improvement.

Yayuan Li, Zhi Cao, Jason J. Corso

Despite the recent strides in video generation, state-of-the-art methods still struggle with elements of visual detail. One particularly challenging case is the class of videos in which the intricate motion of the hand coupled with a mostly stable and otherwise distracting environment is necessary to convey the execution of some complex action and its effects. To address these challenges, we introduce a new method for video generation that focuses on hand-centric actions. Our diffusion-based method incorporates two distinct innovations. First, we propose an automatic method to generate the motion area -- the region in the video in which the detailed activities occur -- guided by both the visual context and the action text prompt, rather than assuming this region can be provided manually as is now commonplace. Second, we introduce a critical Hand Refinement Loss to guide the diffusion model to focus on smooth and consistent hand poses. We evaluate our method on challenging augmented datasets based on EpicKitchens and Ego4D, demonstrating significant improvements over state-of-the-art methods in terms of action clarity, especially of the hand motion in the target region, across diverse environments and actions. Video results can be found in https://excitedbutter.github.io/project_page

Miaowei Wang, Qingxuan Yan, Zhi Cao et al.

Text-guided dynamic 3D character generation has advanced rapidly, yet producing high-quality motion that faithfully reflects rich textual descriptions remains challenging. Existing methods tend to generate limited sub-actions or incoherent motion due to fixed-length temporal inputs and discrete frame-wise representations that fail to capture rich motion semantics. We address these limitations by representing motion with continuous differentiable B-spline curves, enabling more effective motion generation without modifying the capabilities of the underlying generative model. Specifically, our closed-form, Laplacian-regularized B-spline solver efficiently compresses variable-length motion sequences into compact representations with a fixed number of control points. Further, we introduce a normal-fusion strategy for input shape adherence along with correspondence-aware and local-rigidity losses for motion-restoration quality. To train our model, we collate BIMO, a new dataset containing diverse variable-length 3D motion sequences with rich, high-quality text annotations. Extensive evaluations show that our feed-forward framework BiMotion generates more expressive, higher-quality, and better prompt-aligned motions than existing state-of-the-art methods, while also achieving faster generation. Our project page is at: https://wangmiaowei.github.io/BiMotion.github.io/.

Jing Bi, Filippos Bellos, Junjia Guo et al.

Test-time thinking (that is, generating explicit intermediate reasoning chains) is known to boost performance in large language models and has recently shown strong gains for large vision language models (LVLMs). However, despite these promising results, there is still no systematic analysis of how thinking actually affects visual reasoning. We provide the first such analysis with a large scale, controlled comparison of thinking for LVLMs, evaluating ten variants from the InternVL3.5 and Qwen3-VL families on MMMU-val under generous token budgets and multi pass decoding. We show that more thinking is not always better; long chains often yield long wrong trajectories that ignore the image and underperform the same models run in standard instruct mode. A deeper analysis reveals that certain short lookback phrases, which explicitly refer back to the image, are strongly enriched in successful trajectories and correlate with better visual grounding. Building on this insight, we propose uncertainty guided lookback, a training free decoding strategy that combines an uncertainty signal with adaptive lookback prompts and breadth search. Our method improves overall MMMU performance, delivers the largest gains in categories where standard thinking is weak, and outperforms several strong decoding baselines, setting a new state of the art under fixed model families and token budgets. We further show that this decoding strategy generalizes, yielding consistent improvements on five additional benchmarks, including two broad multimodal suites and math focused visual reasoning datasets.

Yayuan Li, Aadit Jain, Filippos Bellos et al.

We introduce Mistake Attribution (MATT), a task for fine-grained understanding of human mistakes in egocentric video. Unlike prior mistake understanding work, which lacks fine-grained output, MATT concretely attributes mistakes to the input instruction text or the attempt video. MATT determines what part of the instruction is violated (semantic role), when the deviation becomes irreversible (the Point-of-No-Return, PNR), and where the mistake appears in the PNR frame. We develop MisEngine, a data engine that automatically constructs attribution-rich mistake samples from existing datasets and inherits their annotations. Applied to large egocentric corpora, MisEngine yields EPIC-KITCHENS-M and Ego4D-M, two datasets that are up to two orders of magnitude larger than prior mistake datasets. We then present MisFormer, a unified attention-based model for mistake attribution across semantic (what), temporal (when), and spatial (where) dimensions, trained using MisEngine supervision. Experiments on our new datasets and prior benchmarks show that MisFormer outperforms strong video-language, temporal localization, hand-object interaction, and mistake-detection baselines.

Yayuan Li, Filippos Bellos, Jason Corso

Current approaches to pose generation rely heavily on intermediate representations, either through two-stage pipelines with quantization or autoregressive models that accumulate errors during inference. This fundamental limitation leads to degraded performance, particularly in long-term pose generation where maintaining temporal coherence is crucial. We propose a novel one-stage architecture that directly generates poses in continuous coordinate space from minimal context - a single RGB image and text description - while maintaining consistent distributions between training and inference. Our key innovation is eliminating the need for intermediate representations or token-based generation by operating directly on pose coordinates through a relative movement prediction mechanism that preserves spatial relationships, and a unified placeholder token approach that enables single-forward generation with identical behavior during training and inference. Through extensive experiments on Penn Action and First-Person Hand Action Benchmark (F-PHAB) datasets, we demonstrate that our approach significantly outperforms existing quantization-based and autoregressive methods, especially in long-term generation scenarios.