Hengyi Wang

Hengyi Wang, Jingwen Wang, Lourdes Agapito · tencent-ai

We present Co-SLAM, a neural RGB-D SLAM system based on a hybrid representation, that performs robust camera tracking and high-fidelity surface reconstruction in real time. Co-SLAM represents the scene as a multi-resolution hash-grid to exploit its high convergence speed and ability to represent high-frequency local features. In addition, Co-SLAM incorporates one-blob encoding, to encourage surface coherence and completion in unobserved areas. This joint parametric-coordinate encoding enables real-time and robust performance by bringing the best of both worlds: fast convergence and surface hole filling. Moreover, our ray sampling strategy allows Co-SLAM to perform global bundle adjustment over all keyframes instead of requiring keyframe selection to maintain a small number of active keyframes as competing neural SLAM approaches do. Experimental results show that Co-SLAM runs at 10-17Hz and achieves state-of-the-art scene reconstruction results, and competitive tracking performance in various datasets and benchmarks (ScanNet, TUM, Replica, Synthetic RGBD). Project page: https://hengyiwang.github.io/projects/CoSLAM

Hengyi Wang, Lourdes Agapito

We present Spann3R, a novel approach for dense 3D reconstruction from ordered or unordered image collections. Built on the DUSt3R paradigm, Spann3R uses a transformer-based architecture to directly regress pointmaps from images without any prior knowledge of the scene or camera parameters. Unlike DUSt3R, which predicts per image-pair pointmaps each expressed in its local coordinate frame, Spann3R can predict per-image pointmaps expressed in a global coordinate system, thus eliminating the need for optimization-based global alignment. The key idea of Spann3R is to manage an external spatial memory that learns to keep track of all previous relevant 3D information. Spann3R then queries this spatial memory to predict the 3D structure of the next frame in a global coordinate system. Taking advantage of DUSt3R's pre-trained weights, and further fine-tuning on a subset of datasets, Spann3R shows competitive performance and generalization ability on various unseen datasets and can process ordered image collections in real time. Project page: \url{https://hengyiwang.github.io/projects/spanner}

Hengyi Wang, Chaoran Zhu, Ziyin Ma et al.

We present methods to estimate the physical properties of household containers and their fillings manipulated by humans. We use a lightweight, pre-trained convolutional neural network with coordinate attention as a backbone model of the pipelines to accurately locate the object of interest and estimate the physical properties in the CORSMAL Containers Manipulation (CCM) dataset. We address the filling type classification with audio data and then combine this information from audio with video modalities to address the filling level classification. For the container capacity, dimension, and mass estimation, we present a data augmentation and consistency measurement to alleviate the over-fitting issue in the CCM dataset caused by the limited number of containers. We augment the training data using an object-of-interest-based re-scaling that increases the variety of physical values of the containers. We then perform the consistency measurement to choose a model with low prediction variance in the same containers under different scenes, which ensures the generalization ability of the model. Our method improves the generalization ability of the models to estimate the property of the containers that were not previously seen in the training.

Ruiqiang Zhang, Hengyi Wang, Chang Liu et al.

Large-scale text-to-image (T2I) diffusion models excel at open-domain synthesis but still struggle with precise text rendering, especially for multi-line layouts, dense typography, and long-tailed scripts such as Chinese. Prior solutions typically require costly retraining or rigid external layout constraints, which can degrade aesthetics and limit flexibility. We propose \textbf{FreeText}, a training-free, plug-and-play framework that improves text rendering by exploiting intrinsic mechanisms of \emph{Diffusion Transformer (DiT)} models. \textbf{FreeText} decomposes the problem into \emph{where to write} and \emph{what to write}. For \emph{where to write}, we localize writing regions by reading token-wise spatial attribution from endogenous image-to-text attention, using sink-like tokens as stable spatial anchors and topology-aware refinement to produce high-confidence masks. For \emph{what to write}, we introduce Spectral-Modulated Glyph Injection (SGMI), which injects a noise-aligned glyph prior with frequency-domain band-pass modulation to strengthen glyph structure and suppress semantic leakage (rendering the concept instead of the word). Extensive experiments on Qwen-Image, FLUX.1-dev, and SD3 variants across longText-Benchmark, CVTG, and our CLT-Bench show consistent gains in text readability while largely preserving semantic alignment and aesthetic quality, with modest inference overhead.

Hengyi Wang, Changjae Oh

We present a refinement framework to boost the performance of pre-trained semi-supervised video object segmentation (VOS) models. Our work is based on scale inconsistency, which is motivated by the observation that existing VOS models generate inconsistent predictions from input frames with different sizes. We use the scale inconsistency as a clue to devise a pixel-level attention module that aggregates the advantages of the predictions from different-size inputs. The scale inconsistency is also used to regularize the training based on a pixel-level variance measured by an uncertainty estimation. We further present a self-supervised online adaptation, tailored for test-time optimization, that bootstraps the predictions without ground-truth masks based on the scale inconsistency. Experiments on DAVIS 16 and DAVIS 17 datasets show that our framework can be generically applied to various VOS models and improve their performance.

Hengyi Wang, Haizhou Shi, Shiwei Tan et al.

Multimodal Large Language Models (MLLMs) have shown significant promise in various applications, leading to broad interest from researchers and practitioners alike. However, a comprehensive evaluation of their long-context capabilities remains underexplored. To address these gaps, we introduce the MultiModal Needle-in-a-haystack (MMNeedle) benchmark, specifically designed to assess the long-context capabilities of MLLMs. Besides multi-image input, we employ image stitching to further increase the input context length, and develop a protocol to automatically generate labels for sub-image level retrieval. Essentially, MMNeedle evaluates MLLMs by stress-testing their capability to locate a target sub-image (needle) within a set of images (haystack) based on textual instructions and descriptions of image contents. This setup necessitates an advanced understanding of extensive visual contexts and effective information retrieval within long-context image inputs. With this benchmark, we evaluate state-of-the-art MLLMs, encompassing both API-based and open-source models. The findings reveal that GPT-4o consistently surpasses other models in long-context scenarios, but suffers from hallucination problems in negative samples, i.e., when needles are not in the haystacks. Our comprehensive long-context evaluation of MLLMs also sheds lights on the considerable performance gap between API-based and open-source models. All the code, data, and instructions required to reproduce the main results are available at https://github.com/Wang-ML-Lab/multimodal-needle-in-a-haystack.

Hangrui Bi, Hengyi Wang, Chence Shi et al.

Reliably predicting the products of chemical reactions presents a fundamental challenge in synthetic chemistry. Existing machine learning approaches typically produce a reaction product by sequentially forming its subparts or intermediate molecules. Such autoregressive methods, however, not only require a pre-defined order for the incremental construction but preclude the use of parallel decoding for efficient computation. To address these issues, we devise a non-autoregressive learning paradigm that predicts reaction in one shot. Leveraging the fact that chemical reactions can be described as a redistribution of electrons in molecules, we formulate a reaction as an arbitrary electron flow and predict it with a novel multi-pointer decoding network. Experiments on the USPTO-MIT dataset show that our approach has established a new state-of-the-art top-1 accuracy and achieves at least 27 times inference speedup over the state-of-the-art methods. Also, our predictions are easier for chemists to interpret owing to predicting the electron flows.

Hangrui Bi, Hengyi Wang, Chence Shi et al.

Reaction prediction is a fundamental problem in computational chemistry. Existing approaches typically generate a chemical reaction by sampling tokens or graph edits sequentially, conditioning on previously generated outputs. These autoregressive generating methods impose an arbitrary ordering of outputs and prevent parallel decoding during inference. We devise a novel decoder that avoids such sequential generating and predicts the reaction in a Non-Autoregressive manner. Inspired by physical-chemistry insights, we represent edge edits in a molecule graph as electron flows, which can then be predicted in parallel. To capture the uncertainty of reactions, we introduce latent variables to generate multi-modal outputs. Following previous works, we evaluate our model on USPTO MIT dataset. Our model achieves both an order of magnitude lower inference latency, with state-of-the-art top-1 accuracy and comparable performance on Top-K sampling.