arXiv subjects

Haohui Chen

Publications and source records attributed to Haohui Chen.

13 recordsLinked to original sources

arXiv2026

SonarLLM: A Native Sonar--Optical Multimodal Large Language Model for Underwater Perception

Reliable underwater perception requires complementary sensing under variable visibility. Optical cameras capture appearance and semantics but degrade rapidly with turbidity, whereas imaging sonar preserves geometry while exhibiting distinct range-azimuth structure and acoustic artifacts. Existing MLLMs, built primarily on optical encoders, are therefore ill-suited to model sonar or adaptively exploit sonar-optical complementarity. We propose SonarLLM, a sonar-optical MLLM that treats sonar as a native perceptual modality. It combines a sonar-specific encoder, modality-specific physics-aware feature enhancement, and reliability-aware hierarchical fusion to align acoustic structure with optical semantics and dynamically adjust their contributions as sensing quality changes. We also introduce SonarBench, a paired benchmark that spans four tasks: recognition, counting, visual question answering, and captioning; and, across the benchmark, three input settings: sonar-only, optical-only, and fusion. By fixing the scene and sonar observation while varying optical degradation, SonarBench enables controlled measurement of cross-modal complementarity. SonarLLM achieves 72.0% macro accuracy across sonar-only recognition, counting, and VQA, outperforming the strongest baseline by 34.4 percentage points, and 68.7% under fusion, exceeding the best baseline by 25.1 points. For recognition and counting, the fusion-over-optical gain grows from 6.0 to 36.0 points as turbidity increases, indicating the increasing complementary value of sonar under controlled optical degradation. Together, these results show that robust heterogeneous perception depends not only on adding sonar, but on representing and weighting it according to its sensing characteristics.

cs.AI↗

arXiv2026

Learning an Interior Layout Policy in a Domain Specific Language Action Space

Indoor scene layout generation is a challenging task in interior design. Existing methods often oversimplify the task by reducing room conditions to coarse 3D bounding boxes and neglecting structural elements such as doors and windows. More fundamentally, many prior approaches formulate spatial reasoning as direct coordinate prediction, thereby casting interior layout design as continuous regression over raw geometric parameters, which hinders the model from learning the underlying reasoning logic of intelligent layout design. We propose \textbf{LayoutDSL}, a novel LLM-based framework for learning an interior layout policy in a domain-specific language (DSL) action space. The DSL provides an explicit symbolic representation of layout information and serves as a structured action space for layout reasoning, where each action corresponds to an interpretable design decision. Under this DSL-based policy learning paradigm, we construct 3D-FrontDSL, a dataset of room-structure annotations paired with synthetic DSL action sequences for supervised fine-tuning. To promote a more generalizable and scalable policy with verifiable feedback, we design rewards grounded in interior design principles and physical plausibility, and optimize the policy via reinforcement learning. Extensive experiments demonstrate that LayoutDSL substantially improves spatial plausibility and design logicality over strong baselines and existing methods.

cs.CV↗

arXiv2026

Home3D 1.0: A High-Fidelity Image-to-3D Asset Generation System for Interior Design

We present Home3D 1.0, a modular image-to-3D generation system that produces high-quality 3D assets from a single reference image, targeting interior design and e-commerce applications. Given a photograph of a furniture or decor item, the system outputs a mesh with physically-based rendering (PBR) materials, and the mesh can be decomposed into material-specific components. The pipeline is organized into four tightly coupled modules: Geometry reconstructs a watertight mesh through latent SDF modelling with a geometry VAE and a coarse-to-fine flow-matching DiT; Texture predicts multiview albedo observations, reprojects them onto the mesh, and completes unseen surface regions with a 3D texture field; Material uses MatWeaver to obtain component masks through video-based segmentation and UV-space voting, then retrieves and bakes PBR maps from a curated material library through hierarchical multi-modal matching; and Parts generates material-editable semantic part meshes with a PartVAE and PartDiT, decoding multi-head part-specific SDF fields in one pass. Each module is evaluated independently with dedicated metrics, highlighting both the current system capability and the remaining gaps toward broader deployment.

cs.CV↗

arXiv2025

Mitigating Estimation Bias with Representation Learning in TD Error-Driven Regularization

Deterministic policy gradient algorithms for continuous control suffer from value estimation biases that degrade performance. While double critics reduce such biases, the exploration potential of double actors remains underexplored. Building on temporal-difference error-driven regularization (TDDR), a double actor-critic framework, this work introduces enhanced methods to achieve flexible bias control and stronger representation learning. We propose three convex combination strategies, symmetric and asymmetric, that balance pessimistic estimates to mitigate overestimation and optimistic exploration via double actors to alleviate underestimation. A single hyperparameter governs this mechanism, enabling tunable control across the bias spectrum. To further improve performance, we integrate augmented state and action representations into the actor and critic networks. Extensive experiments show that our approach consistently outperforms benchmarks, demonstrating the value of tunable bias and revealing that both overestimation and underestimation can be exploited differently depending on the environment.

cs.LG↗

arXiv2025

Mildly Conservative Regularized Evaluation for Offline Reinforcement Learning

Offline reinforcement learning (RL) seeks to learn optimal policies from static datasets without further environment interaction. A key challenge is the distribution shift between the learned and behavior policies, leading to out-of-distribution (OOD) actions and overestimation. To prevent gross overestimation, the value function must remain conservative; however, excessive conservatism may hinder performance improvement. To address this, we propose the mildly conservative regularized evaluation (MCRE) framework, which balances conservatism and performance by combining temporal difference (TD) error with a behavior cloning term in the Bellman backup. Building on this, we develop the mildly conservative regularized Q-learning (MCRQ) algorithm, which integrates MCRE into an off-policy actor-critic framework. Experiments show that MCRQ outperforms strong baselines and state-of-the-art offline RL algorithms on benchmark datasets.

cs.LG↗

arXiv2024

Double Actor-Critic with TD Error-Driven Regularization in Reinforcement Learning

To obtain better value estimation in reinforcement learning, we propose a novel algorithm based on the double actor-critic framework with temporal difference error-driven regularization, abbreviated as TDDR. TDDR employs double actors, with each actor paired with a critic, thereby fully leveraging the advantages of double critics. Additionally, TDDR introduces an innovative critic regularization architecture. Compared to classical deterministic policy gradient-based algorithms that lack a double actor-critic structure, TDDR provides superior estimation. Moreover, unlike existing algorithms with double actor-critic frameworks, TDDR does not introduce any additional hyperparameters, significantly simplifying the design and implementation process. Experiments demonstrate that TDDR exhibits strong competitiveness compared to benchmark algorithms in challenging continuous control tasks.

cs.LG↗

arXiv2021

A Histogram Thresholding Improvement to Mask R-CNN for Scalable Segmentation of New and Old Rural Buildings

Mapping new and old buildings are of great significance for understanding socio-economic development in rural areas. In recent years, deep neural networks have achieved remarkable building segmentation results in high-resolution remote sensing images. However, the scarce training data and the varying geographical environments have posed challenges for scalable building segmentation. This study proposes a novel framework based on Mask R-CNN, named HTMask R-CNN, to extract new and old rural buildings even when the label is scarce. The framework adopts the result of single-object instance segmentation from the orthodox Mask R-CNN. Further, it classifies the rural buildings into new and old ones based on a dynamic grayscale threshold inferred from the result of a two-object instance segmentation task where training data is scarce. We found that the framework can extract more buildings and achieve a much higher mean Average Precision (mAP) than the orthodox Mask R-CNN model. We tested the novel framework's performance with increasing training data and found that it converged even when the training samples were limited. This framework's main contribution is to allow scalable segmentation by using significantly fewer training samples than traditional machine learning practices. That makes mapping China's new and old rural buildings viable.

cs.CV↗

arXiv2019

Origin-Destination Flow Maps in Immersive Environments

Immersive virtual- and augmented-reality headsets can overlay a flat image against any surface or hang virtual objects in the space around the user. The technology is rapidly improving and may, in the long term, replace traditional flat panel displays in many situations. When displays are no longer intrinsically flat, how should we use the space around the user for abstract data visualisation? In this paper, we ask this question with respect to origin-destination flow data in a global geographic context. We report on the findings of three studies exploring different spatial encodings for flow maps. The first experiment focuses on different 2D and 3D encodings for flows on flat maps. We find that participants are significantly more accurate with raised flow paths whose height is proportional to flow distance but fastest with traditional straight line 2D flows. In our second and third experiment, we compared flat maps, 3D globes and a novel interactive design we call MapsLink, involving a pair of linked flat maps. We find that participants took significantly more time with MapsLink than other flow maps while the 3D globe with raised flows was the fastest, most accurate, and most preferred method. Our work suggests that careful use of the third spatial dimension can resolve visual clutter in complex flow maps.

cs.HC↗

arXiv2019

Maps and Globes in Virtual Reality

This paper explores different ways to render world-wide geographic maps in virtual reality (VR). We compare: (a) a 3D exocentric globe, where the user's viewpoint is outside the globe; (b) a flat map (rendered to a plane in VR); (c) an egocentric 3D globe, with the viewpoint inside the globe; and (d) a curved map, created by projecting the map onto a section of a sphere which curves around the user. In all four visualisations the geographic centre can be smoothly adjusted with a standard handheld VR controller and the user, through a head-tracked headset, can physically move around the visualisation. For distance comparison, exocentric globe is more accurate than egocentric globe and flat map. For area comparison, more time is required with exocentric and egocentric globes than with flat and curved maps. For direction estimation, the exocentric globe is more accurate and faster than the other visual presentations. Our study participants had a weak preference for the exocentric globe. Generally, the curved map had benefits over the flat map. In almost all cases the egocentric globe was found to be the least effective visualisation. Overall, our results provide support for the use of exocentric globes for geographic visualisation in mixed-reality.

cs.HC↗

arXiv2017

Weather impacts expressed sentiment

We conduct the largest ever investigation into the relationship between meteorological conditions and the sentiment of human expressions. To do this, we employ over three and a half billion social media posts from tens of millions of individuals from both Facebook and Twitter between 2009 and 2016. We find that cold temperatures, hot temperatures, precipitation, narrower daily temperature ranges, humidity, and cloud cover are all associated with worsened expressions of sentiment, even when excluding weather-related posts. We compare the magnitude of our estimates with the effect sizes associated with notable historical events occurring within our data.

stat.AP↗

arXiv2017

Investigating the potential of social network data for transport demand models

Location-based social network data offers the promise of collecting the data from a large base of users over a longer span of time at negligible cost. While several studies have applied social network data to activity and mobility analysis, a comparison with travel diaries and general statistics has been lacking. In this paper, we analysed geo-referenced Twitter activities from a large number of users in Singapore and neighbouring countries. By combining this data, population statistics and travel diaries and applying clustering techniques, we addressed detection of activity locations, as well as spatial separation and transitions between these locations. Kernel density estimation performs best to detect activity locations due to the scattered nature of the twitter data; more activity locations are detected per user than reported in the travel survey. The descriptive analysis shows that determining home locations is more difficult than detecting work locations for most planning zones. Spatial separations between detected activity locations from Twitter data - as reported in a travel survey and captured by public transport smart card data - are mostly similarly distributed, but also show relevant differences for very short and very long distances. This also holds for the transitions between zones. Whether the differences between Twitter data and other data sources stem from differences in the population sub-sample, clustering methodology, or whether social networks are being used significantly more at specific locations must be determined by further research. Despite these shortcomings, location-based social network data offers a promising data source for insights into activity locations and mobility patterns, especially for regions where travel survey data is not readily available.

cs.SI↗

arXiv2015

Nowcasting Disaster Damage

Could social media data aid in disaster response and damage assessment? Countries face both an increasing frequency and intensity of natural disasters due to climate change. And during such events, citizens are turning to social media platforms for disaster-related communication and information. Social media improves situational awareness, facilitates dissemination of emergency information, enables early warning systems, and helps coordinate relief efforts. Additionally, spatiotemporal distribution of disaster-related messages helps with real-time monitoring and assessment of the disaster itself. Here we present a multiscale analysis of Twitter activity before, during, and after Hurricane Sandy. We examine the online response of 50 metropolitan areas of the United States and find a strong relationship between proximity to Sandy's path and hurricane-related social media activity. We show that real and perceived threats -- together with the physical disaster effects -- are directly observable through the intensity and composition of Twitter's message stream. We demonstrate that per-capita Twitter activity strongly correlates with the per-capita economic damage inflicted by the hurricane. Our findings suggest that massive online social networks can be used for rapid assessment ("nowcasting") of damage caused by a large-scale disaster.

cs.SI↗

arXiv2014

Performance of Social Network Sensors During Hurricane Sandy

Information flow during catastrophic events is a critical aspect of disaster management. Modern communication platforms, in particular online social networks, provide an opportunity to study such flow, and a mean to derive early-warning sensors, improving emergency preparedness and response. Performance of the social networks sensor method, based on topological and behavioural properties derived from the "friendship paradox", is studied here for over 50 million Twitter messages posted before, during, and after Hurricane Sandy. We find that differences in user's network centrality effectively translate into moderate awareness advantage (up to 26 hours); and that geo-location of users within or outside of the hurricane-affected area plays significant role in determining the scale of such advantage. Emotional response appears to be universal regardless of the position in the network topology, and displays characteristic, easily detectable patterns, opening a possibility of implementing a simple "sentiment sensing" technique to detect and locate disasters.

cs.SI↗