Jia-Fong Yeh

Chi-Ming Chung, Yang-Che Tseng, Ya-Ching Hsu et al.

A spatial AI that can perform complex tasks through visual signals and cooperate with humans is highly anticipated. To achieve this, we need a visual SLAM that easily adapts to new scenes without pre-training and generates dense maps for downstream tasks in real-time. None of the previous learning-based and non-learning-based visual SLAMs satisfy all needs due to the intrinsic limitations of their components. In this work, we develop a visual SLAM named Orbeez-SLAM, which successfully collaborates with implicit neural representation and visual odometry to achieve our goals. Moreover, Orbeez-SLAM can work with the monocular camera since it only needs RGB inputs, making it widely applicable to the real world. Results show that our SLAM is up to 800x faster than the strong baseline with superior rendering outcomes. Code link: https://github.com/MarvinChung/Orbeez-SLAM.

Gueter Josmy Faure, Min-Hung Chen, Jia-Fong Yeh et al.

Vision-Language Models (VLMs) have demonstrated remarkable capabilities in general video understanding, yet they often struggle with the fine-grained comprehension crucial for real-world applications requiring nuanced interpretation of human actions and interactions. While some recent human-centric benchmarks evaluate aspects of model behaviour such as fairness/ethics, emotion perception, and broader human-centric metrics, they do not combine long-form videos, very dense QA coverage, and frame-level spatial/temporal grounding at scale. To bridge this gap, we introduce FineBench, a human-centric video question answering (VQA) benchmark specifically designed to assess fine-grained understanding. FineBench comprises 199,420 multiple-choice QA pairs densely annotated across 64 long-form videos (15 minutes each), focusing on detailed person movement, person interaction, and object manipulation, including compositional actions. Our extensive evaluation reveals that while proprietary models like GPT-5 achieve respectable performance, current open-source VLMs significantly underperform, struggling particularly with spatial reasoning in multi-person scenes and distinguishing subtle differences in human movements and interactions. To address these identified weaknesses, we propose FineAgent, a modular framework that enhances VLMs by leveraging a Localizer and a Descriptor. Experiments show that FineAgent consistently improves the performance of various open VLMs on FineBench. FineBench provides a rigorous testbed for future research into fine-grained human-centric video understanding, while FineAgent offers a practical approach to enhance such reasoning in current VLMs.

Ru-Fen Jheng, Tsung-Han Wu, Jia-Fong Yeh et al.

Nowadays, the need for user editing in a 3D scene has rapidly increased due to the development of AR and VR technology. However, the existing 3D scene completion task (and datasets) cannot suit the need because the missing regions in scenes are generated by the sensor limitation or object occlusion. Thus, we present a novel task named free-form 3D scene inpainting. Unlike scenes in previous 3D completion datasets preserving most of the main structures and hints of detailed shapes around missing regions, the proposed inpainting dataset, FF-Matterport, contains large and diverse missing regions formed by our free-form 3D mask generation algorithm that can mimic human drawing trajectories in 3D space. Moreover, prior 3D completion methods cannot perform well on this challenging yet practical task, simply interpolating nearby geometry and color context. Thus, a tailored dual-stream GAN method is proposed. First, our dual-stream generator, fusing both geometry and color information, produces distinct semantic boundaries and solves the interpolation issue. To further enhance the details, our lightweight dual-stream discriminator regularizes the geometry and color edges of the predicted scenes to be realistic and sharp. We conducted experiments with the proposed FF-Matterport dataset. Qualitative and quantitative results validate the superiority of our approach over existing scene completion methods and the efficacy of all proposed components.

Hsin-Ying Lee, Hung-Ting Su, Bing-Chen Tsai et al.

While recent large-scale video-language pre-training made great progress in video question answering, the design of spatial modeling of video-language models is less fine-grained than that of image-language models; existing practices of temporal modeling also suffer from weak and noisy alignment between modalities. To learn fine-grained visual understanding, we decouple spatial-temporal modeling and propose a hybrid pipeline, Decoupled Spatial-Temporal Encoders, integrating an image- and a video-language encoder. The former encodes spatial semantics from larger but sparsely sampled frames independently of time, while the latter models temporal dynamics at lower spatial but higher temporal resolution. To help the video-language model learn temporal relations for video QA, we propose a novel pre-training objective, Temporal Referring Modeling, which requires the model to identify temporal positions of events in video sequences. Extensive experiments demonstrate that our model outperforms previous work pre-trained on orders of magnitude larger datasets.

Yi-Syuan Liou, Tsung-Han Wu, Jia-Fong Yeh et al.

Obtaining large-scale labeled object detection dataset can be costly and time-consuming, as it involves annotating images with bounding boxes and class labels. Thus, some specialized active learning methods have been proposed to reduce the cost by selecting either coarse-grained samples or fine-grained instances from unlabeled data for labeling. However, the former approaches suffer from redundant labeling, while the latter methods generally lead to training instability and sampling bias. To address these challenges, we propose a novel approach called Multi-scale Region-based Active Learning (MuRAL) for object detection. MuRAL identifies informative regions of various scales to reduce annotation costs for well-learned objects and improve training performance. The informative region score is designed to consider both the predicted confidence of instances and the distribution of each object category, enabling our method to focus more on difficult-to-detect classes. Moreover, MuRAL employs a scale-aware selection strategy that ensures diverse regions are selected from different scales for labeling and downstream finetuning, which enhances training stability. Our proposed method surpasses all existing coarse-grained and fine-grained baselines on Cityscapes and MS COCO datasets, and demonstrates significant improvement in difficult category performance.

Gueter Josmy Faure, Jia-Fong Yeh, Min-Hung Chen et al.

Long-form video understanding presents unique challenges that extend beyond traditional short-video analysis approaches, particularly in capturing long-range dependencies, processing redundant information efficiently, and extracting high-level semantic concepts. To address these challenges, we propose a novel approach that more accurately reflects human cognition. This paper introduces HERMES: temporal-coHERent long-forM understanding with Episodes and Semantics, featuring two versatile modules that can enhance existing video-language models or operate as a standalone system. Our Episodic COmpressor (ECO) efficiently aggregates representations from micro to semi-macro levels, reducing computational overhead while preserving temporal dependencies. Our Semantics ReTRiever (SeTR) enriches these representations with semantic information by focusing on broader context, dramatically reducing feature dimensionality while preserving relevant macro-level information. We demonstrate that these modules can be seamlessly integrated into existing SOTA models, consistently improving their performance while reducing inference latency by up to 43% and memory usage by 46%. As a standalone system, HERMES achieves state-of-the-art performance across multiple long-video understanding benchmarks in both zero-shot and fully-supervised settings.

Hung-Ting Su, Ting-Jun Wang, Jia-Fong Yeh et al.

Conventional Vision-and-Language Navigation (VLN) benchmarks assume instructions are feasible and the referenced target exists, leaving agents ill-equipped to handle false-premise goals. We introduce VLN-NF, a benchmark with false-premise instructions where the target is absent from the specified room and agents must navigate, gather evidence through in-room exploration, and explicitly output NOT-FOUND. VLN-NF is constructed via a scalable pipeline that rewrites VLN instructions using an LLM and verifies target absence with a VLM, producing plausible yet factually incorrect goals. We further propose REV-SPL to jointly evaluate room reaching, exploration coverage, and decision correctness. To address this challenge, we present ROAM, a two-stage hybrid that combines supervised room-level navigation with LLM/VLM-driven in-room exploration guided by a free-space clearance prior. ROAM achieves the best REV-SPL among compared methods, while baselines often under-explore and terminate prematurely under unreliable instructions. VLN-NF project page can be found at https://vln-nf.github.io/.

Pei-An Chen, Yong-Ching Liang, Jia-Fong Yeh et al.

Intelligent embodied agents should not simply follow instructions, as real-world environments often involve unexpected conditions and exceptions. However, existing methods usually focus on directly executing instructions, without considering whether the target objects can actually be manipulated, meaning they fail to assess available affordances. To address this limitation, we introduce DynAfford, a benchmark that evaluates embodied agents in dynamic environments where object affordances may change over time and are not specified in the instruction. DynAfford requires agents to perceive object states, infer implicit preconditions, and adapt their actions accordingly. To enable this capability, we introduce ADAPT, a plug-and-play module that augments existing planners with explicit affordance reasoning. Experiments demonstrate that incorporating ADAPT significantly improves robustness and task success across both seen and unseen environments. We also show that a domain-adapted, LoRA-finetuned vision-language model used as the affordance inference backend outperforms a commercial LLM (GPT-4o), highlighting the importance of task-aligned affordance grounding.

Huang-Yu Chen, Jia-Fong Yeh, Jia-Wei Liao et al.

LiDAR-based 3D object detection is a critical technology for the development of autonomous driving and robotics. However, the high cost of data annotation limits its advancement. We propose a novel and effective active learning (AL) method called Distribution Discrepancy and Feature Heterogeneity (DDFH), which simultaneously considers geometric features and model embeddings, assessing information from both the instance-level and frame-level perspectives. Distribution Discrepancy evaluates the difference and novelty of instances within the unlabeled and labeled distributions, enabling the model to learn efficiently with limited data. Feature Heterogeneity ensures the heterogeneity of intra-frame instance features, maintaining feature diversity while avoiding redundant or similar instances, thus minimizing annotation costs. Finally, multiple indicators are efficiently aggregated using Quantile Transform, providing a unified measure of informativeness. Extensive experiments demonstrate that DDFH outperforms the current state-of-the-art (SOTA) methods on the KITTI and Waymo datasets, effectively reducing the bounding box annotation cost by 56.3% and showing robustness when working with both one-stage and two-stage models.

Tzu-Jung Lin, Jia-Fong Yeh, Hung-Ting Su et al.

In open-vocabulary mobile manipulation (OVMM), task success often hinges on the selection of an appropriate base placement for the robot. Existing approaches typically navigate to proximity-based regions without considering affordances, resulting in frequent manipulation failures. We propose Affordance-Guided Coarse-to-Fine Exploration, a zero-shot framework for base placement that integrates semantic understanding from vision-language models (VLMs) with geometric feasibility through an iterative optimization process. Our method constructs cross-modal representations, namely Affordance RGB and Obstacle Map+, to align semantics with spatial context. This enables reasoning that extends beyond the egocentric limitations of RGB perception. To ensure interaction is guided by task-relevant affordances, we leverage coarse semantic priors from VLMs to guide the search toward task-relevant regions and refine placements with geometric constraints, thereby reducing the risk of convergence to local optima. Evaluated on five diverse open-vocabulary mobile manipulation tasks, our system achieves an 85% success rate, significantly outperforming classical geometric planners and VLM-based methods. This demonstrates the promise of affordance-aware and multimodal reasoning for generalizable, instruction-conditioned planning in OVMM.

Ting-Kang Yen, Igor Morawski, Shusil Dangi et al.

Event-based object detection has recently garnered attention in the computer vision community due to the exceptional properties of event cameras, such as high dynamic range and no motion blur. However, feature asynchronism and sparsity cause invisible objects due to no relative motion to the camera, posing a significant challenge in the task. Prior works have studied various implicit-learned memories to retain as many temporal cues as possible. However, implicit memories still struggle to preserve long-term features effectively. In this paper, we consider those invisible objects as pseudo-occluded objects and aim to detect them by tracking through occlusions. Firstly, we introduce the visibility attribute of objects and contribute an auto-labeling algorithm to not only clean the existing event camera dataset but also append additional visibility labels to it. Secondly, we exploit tracking strategies for pseudo-occluded objects to maintain their permanence and retain their bounding boxes, even when features have not been available for a very long time. These strategies can be treated as an explicit-learned memory guided by the tracking objective to record the displacements of objects across frames. Lastly, we propose a spatio-temporal feature aggregation module to enrich the latent features and a consistency loss to increase the robustness of the overall pipeline. We conduct comprehensive experiments to verify our method's effectiveness where still objects are retained, but real occluded objects are discarded. The results demonstrate that (1) the additional visibility labels can assist in supervised training, and (2) our method outperforms state-of-the-art approaches with a significant improvement of 7.9% absolute mAP.

Gueter Josmy Faure, Min-Hung Chen, Jia-Fong Yeh et al.

This paper introduces MovieCORE, a novel video question answering (VQA) dataset designed to probe deeper cognitive understanding of movie content. Unlike existing datasets that focus on surface-level comprehension, MovieCORE emphasizes questions that engage System-2 thinking while remaining specific to the video material. We present an innovative agentic brainstorming approach, utilizing multiple large language models (LLMs) as thought agents to generate and refine high-quality question-answer pairs. To evaluate dataset quality, we develop a set of cognitive tests assessing depth, thought-provocation potential, and syntactic complexity. We also propose a comprehensive evaluation scheme for assessing VQA model performance on deeper cognitive tasks. To address the limitations of existing video-language models (VLMs), we introduce an agentic enhancement module, Agentic Choice Enhancement (ACE), which improves model reasoning capabilities post-training by up to 25%. Our work contributes to advancing movie understanding in AI systems and provides valuable insights into the capabilities and limitations of current VQA models when faced with more challenging, nuanced questions about cinematic content. Our project page, dataset and code can be found at https://joslefaure.github.io/assets/html/moviecore.html.

Zhao Wang, Briti Gangopadhyay, Jia-Fong Yeh et al.

Recent advancements in offline Reinforcement Learning (Offline RL) have led to an increased focus on methods based on conservative policy updates to address the Out-of-Distribution (OOD) issue. These methods typically involve adding behavior regularization or modifying the critic learning objective, focusing primarily on states or actions with substantial dataset support. However, we challenge this prevailing notion by asserting that the absence of an action or state from a dataset does not necessarily imply its suboptimality. In this paper, we propose a novel approach to tackle the OOD problem. We introduce an offline RL teacher-student framework, complemented by a policy similarity measure. This framework enables the student policy to gain insights not only from the offline RL dataset but also from the knowledge transferred by a teacher policy. The teacher policy is trained using another dataset consisting of state-action pairs, which can be viewed as practical domain knowledge acquired without direct interaction with the environment. We believe this additional knowledge is key to effectively solving the OOD issue. This research represents a significant advancement in integrating a teacher-student network into the actor-critic framework, opening new avenues for studies on knowledge transfer in offline RL and effectively addressing the OOD challenge.

Briti Gangopadhyay, Zhao Wang, Jia-Fong Yeh et al.

With the ability to learn from static datasets, Offline Reinforcement Learning (RL) emerges as a compelling avenue for real-world applications. However, state-of-the-art offline RL algorithms perform sub-optimally when confronted with limited data confined to specific regions within the state space. The performance degradation is attributed to the inability of offline RL algorithms to learn appropriate actions for rare or unseen observations. This paper proposes a novel domain knowledge-based regularization technique and adaptively refines the initial domain knowledge to considerably boost performance in limited data with partially omitted states. The key insight is that the regularization term mitigates erroneous actions for sparse samples and unobserved states covered by domain knowledge. Empirical evaluations on standard discrete environment datasets demonstrate a substantial average performance increase of at least 27% compared to existing offline RL algorithms operating on limited data.

Po-Shao Lin, Jia-Fong Yeh, Yi-Ting Chen et al.

We observe that current state-of-the-art (SOTA) methods suffer from the performance imbalance issue when performing multi-task reinforcement learning (MTRL) tasks. While these methods may achieve impressive performance on average, they perform extremely poorly on a few tasks. To address this, we propose a new and effective method called STARS, which consists of two novel strategies: a shared-unique feature extractor and task-aware prioritized sampling. First, the shared-unique feature extractor learns both shared and task-specific features to enable better synergy of knowledge between different tasks. Second, the task-aware sampling strategy is combined with the prioritized experience replay for efficient learning on tasks with poor performance. The effectiveness and stability of our STARS are verified through experiments on the mainstream Meta-World benchmark. From the results, our STARS statistically outperforms current SOTA methods and alleviates the performance imbalance issue. Besides, we visualize the learned features to support our claims and enhance the interpretability of STARS.

Jia-Fong Yeh, Chi-Ming Chung, Hung-Ting Su et al.

In few-shot imitation learning (FSIL), using behavioral cloning (BC) to solve unseen tasks with few expert demonstrations becomes a popular research direction. The following capabilities are essential in robotics applications: (1) Behaving in compound tasks that contain multiple stages. (2) Retrieving knowledge from few length-variant and misalignment demonstrations. (3) Learning from a different expert. No previous work can achieve these abilities at the same time. In this work, we conduct FSIL problem under the union of above settings and introduce a novel stage conscious attention network (SCAN) to retrieve knowledge from few demonstrations simultaneously. SCAN uses an attention module to identify each stage in length-variant demonstrations. Moreover, it is designed under demonstration-conditioned policy that learns the relationship between experts and agents. Experiment results show that SCAN can learn from different experts without fine-tuning and outperform baselines in complicated compound tasks with explainable visualization.

Tung-I Chen, Yueh-Cheng Liu, Hung-Ting Su et al.

While recent progress has significantly boosted few-shot classification (FSC) performance, few-shot object detection (FSOD) remains challenging for modern learning systems. Existing FSOD systems follow FSC approaches, ignoring critical issues such as spatial variability and uncertain representations, and consequently result in low performance. Observing this, we propose a novel \textbf{Dual-Awareness Attention (DAnA)} mechanism that enables networks to adaptively interpret the given support images. DAnA transforms support images into \textbf{query-position-aware} (QPA) features, guiding detection networks precisely by assigning customized support information to each local region of the query. In addition, the proposed DAnA component is flexible and adaptable to multiple existing object detection frameworks. By adopting DAnA, conventional object detection networks, Faster R-CNN and RetinaNet, which are not designed explicitly for few-shot learning, reach state-of-the-art performance in FSOD tasks. In comparison with previous methods, our model significantly increases the performance by 47\% (+6.9 AP), showing remarkable ability under various evaluation settings.

Jia-Fong Yeh, Hsin-Ying Lee, Bing-Chen Tsai et al.

In recent years, few-shot learning problems have received a lot of attention. While methods in most previous works were trained and tested on datasets in one single domain, cross-domain few-shot learning is a brand-new branch of few-shot learning problems, where models handle datasets in different domains between training and testing phases. In this paper, to solve the problem that the model is pre-trained (meta-trained) on a single dataset while fine-tuned on datasets in four different domains, including common objects, satellite images, and medical images, we propose a novel large margin fine-tuning method (LMM-PQS), which generates pseudo query images from support images and fine-tunes the feature extraction modules with a large margin mechanism inspired by methods in face recognition. According to the experiment results, LMM-PQS surpasses the baseline models by a significant margin and demonstrates that our approach is robust and can easily adapt pre-trained models to new domains with few data.