Long Lan

Research

Research directions

Visual perception of non-cooperative targets is our central task. Open-environment learning and on-device embodied intelligence provide supporting methods. Each direction lists selected publications and related foundational work.

Visual Perception of Non-Cooperative Targets

Target discovery, localization, and tracking in complex real-world environments.

1.1Wide-Area Target Discovery (Remote Sensing Detection and Cross-Platform Matching)

The XLRS-Bench hierarchy of remote-sensing perception and reasoning capabilities and subtasks.
XLRS-Bench evaluates multimodal models across perception, reasoning, and sixteen fine-grained remote-sensing subtasks.Figure 3 from Wang et al., XLRS-Bench, CVPR 2025. Cropped from the official open-access paper; Long Lan is a coauthor. ↗

1.2Long-Term Target Tracking under Occlusion

Overview of a UAV multi-object tracking pipeline with temporal feature aggregation, detection, re-identification, and association.
UAV multi-object tracking with temporal feature aggregation and topology-integrated embeddings.Figure 2 from Xiao et al., Drones 7(6), 389 (2023), CC BY 4.0. Resized; Long Lan is a coauthor. ↗

Machine Learning in Open Environments

Foundational learning methods that support reliable visual perception in open environments.

2.1Out-of-Distribution Detection

Examples of fair, biased, and fairness-confused out-of-distribution detection.
Fairness confusion can cause sensitive attributes and feature shifts to distort out-of-distribution decisions.Figure 1 from Hu et al., Neural Networks 205, 109564 (2027; online 2026); Long Lan is a coauthor. ↗

2.5Learning with Noisy Labels

Parameter-space trajectories for additive decomposition during learning with noisy labels.
TNLPAD separates parameters associated with clean and mislabeled samples over the course of training.Figure 3 from Wang et al., TPAMI 46(9), 6341–6354 (2024); Long Lan is a coauthor. ↗

On-Device Embodied Intelligence

Efficient models and active perception as enabling technologies for visual tasks.

3.1Model Compression through Knowledge Distillation

Knowledge-distillation framework that enhances the magnitude and diversity of non-target classes.
Knowledge distillation through complementary magnitude and diversity enhancement.Figure 2 from Li et al., Sensors 24(11), 3617 (2024), CC BY 4.0. Resized; Long Lan is a coauthor. ↗

3.4Behavior Prediction and Self-Verification with World Models

Related foundations for sequential prediction and goal inference in multi-agent settings.

Mamba-CDSP framework for time-varying counterfactual prediction from historical observations.
Mamba-CDSP models sequential observations and debiases state-space representations for counterfactual prediction.Figure 3 from Wang et al., ICLR 2025; Long Lan is a coauthor. Cropped from the author-hosted paper. ↗