张靖东
Personal Website
I am currently a fourth-year Ph.D. student majoring in Computer Science at Texas A&M University, advised by Prof. Wenping Wang and Prof. Xin Li.
I spent several wonderful summers interning in industry. In summer 2026, I was a research intern at NVIDIA Research, working with Chris Choy. I also collaborated with Dr. Xiaohang Zhan and Lingzhi Zhang at Adobe.
I received my Bachelor of Engineering degree at Fudan University in 2023. I also worked as a research assistant at HKUST CSE with Prof. Dan Xu. Previously, I worked with Profs. Tao Chen and Jiayuan Fan.
My research builds scalable foundation models for spatial intelligence, with interests in:
Research Intern, working on foundational spatial understanding.
May 2026 - Aug 2026
Research Intern, working on generative soft inpainting.
May 2025 - Aug 2025
Research Intern, working on high-quality 3D asset generation.
May 2024 - Aug 2024
Overview: A benchmark for how well frontier agents improve themselves: given a task, the agent iteratively refines its own solution, tools and prompts to raise its score within a fixed budget. Tasks span visual perception and reasoning, robot control, 3D reconstruction and more, and we are actively adding tasks, tracks and models.
Abstract: We present S4VY, a Segment Anything model that establishes object identities through shared 4D visual geometry instead of propagating 2D masks along ordered frames: i) Feed-forward Segment Anything in 4D: persistent object queries turn shared visual-geometric features into exhaustive class-agnostic 4D instances, each keeping one identity across all observations, and any of them can be selected by point or box prompts, ii) Agentic language grounding: a dual-stream grounder, active tree search and a comparative critic locate natural-language targets across large dynamic observation sets.
Abstract: We propose Imagining in 360, a framework for Humanoid Visual Search (HVS) in immersive 360 environments with: i) Decoupled paradigm: separates intuitive spatial imagination from action planning for more grounded exploration, ii) Probabilistic Imaginator: learns spatial priors and guides the Actor via probabilistic estimations, iii) Scalable data engine: a fully automated pipeline producing 1.92M training samples without manual trajectory labels.
Abstract: We propose a foundational panoramic scene understanding model capabale of predicting multiple dense prediction tasks. We achieve this by i) curating large dataset with auto labeling pipeline by existing perspective prediction models, ii) propose PD-BridgeNet to tackle the multi-task interaction challenges under EPR distortions .
Abstract: We promote clipped LDR panoramas to linear HDRIs for image-based lighting: i) a fine-tuned video diffusion model reconstructs the saturated regions, ii) rendering-space losses — a diffuse spherical-harmonic loss and a glossy GGX loss — directly supervise the lighting behavior of the result, recovering light intensity, color and direction.
Abstract: We propose a foundational image soft effect removal (SER) model with: i) a large, curated pair-wise dataset with diverse soft effects (e.g. lens flare, haze, shadows, and reflections), ii) fine-grained user control with spatial masks and strength control, iii) generalize on zero-shot unseen effects, iv) add or enhance effects.
Abstract: SPGen leverages Spherical Projection (SP) to generate high-quality 3D shapes with i) Consistency: SP maps ensure view-consistent and unambiguous 3D reconstruction, ii) Flexibility: Supports arbitrary topologies, iii) Efficiency: Inherit powerful 2D diffusion priors and enables efficient finetuning.
Abstract: We present SolidGS, which reconstructs a consolidated Gaussian field from sparse inputs. Given only three input views, our approach enables high-precision and detailed mesh extraction, and high-quality novel view synthesis, achieved within just three minutes.
Abstract: This research proposes a new approach to multi-task dense predictions with partially labeled data. We introduce hierarchical task tokens (HiTTs) to capture multi-level representations. The global task tokens conduct cross-task interactions and transfer knowledge from labeled to unlabeled tasks.
Abstract: This work introduces a novel BridgeNet for multi-task learning on dense predictions. It uses a Bridge Feature Extractor (BFE) to create strong bridge features and a Task Pattern Propagation (TPP) to solve the task-pattern entanglement issue, resulting in task-specific features with higher quality.
Abstract: We introduce a new approach for cross-domain pedestrian one-stage detectors. The paper identifies a foreground-background misalignment issue in image-level feature alignment, and a novel framework, Background-Focused Distribution Alignment (BFDA) is proposed to address this issue.
Department of Computer Science and Engineering
Ph.D. Student
August 2023 - Present
Intelligent Science and Technology (Excellent Class)
Undergraduate Degree
September 2019 - June 2023CVPR, ICCV, ECCV, NeurIPS, ICLR, ICRA, AAAI, TPAMI, TVCG
I love photography and road trips. Intermediate skier.
