Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Videos

Optical Flow as Spatial-Temporal Attention Learners.

Yawen Lu, Cheng Han, Qifan Wang

IEEE Transactions on Pattern Analysis and Machine Intelligence

|October 3, 2024

Summary

This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Long-distance lanthanide migration regulated by interfacial lattice strain in nanostructures.

Nature communications·2026

Same author

High-Efficiency Production of <i>N</i>-Acetyllactosamine (LacNAc) Using a Cell-Coupled Biocatalytic Strategy with Engineered <i>Escherichia coli</i> and <i>Saccharomyces cerevisiae</i>.

ACS synthetic biology·2026

Same author

Development and application of an intelligent assessment system for medical clinical skill training.

NPJ digital medicine·2026

Same author

Cellular Coupling Fermentation Utilizing Engineered <i>Escherichia coli</i> and Yeast for Efficient Biosynthesis of Sialyllacto-N-tetraose a via Modular Strategy.

Journal of agricultural and food chemistry·2026

Same author

Engineering phage endolysins and receptor-binding proteins for foodborne pathogen control and detection: A review and AI-driven framework.

International journal of food microbiology·2026

Same author

A Wireless Battery-Free Probe-Free Disposable Electrical-Digital-PCR Chip.

IEEE transactions on biomedical circuits and systems·2026

Same journal

Relation DETR+: Exploring Explicit Position Relation Prior for Dense Prediction.

IEEE transactions on pattern analysis and machine intelligence·2026

Same journal

RBF++: Quantifying and Optimizing Reasoning Boundaries across Measurable and Unmeasurable Capabilities for Chain-of-Thought Reasoning.

IEEE transactions on pattern analysis and machine intelligence·2026

Same journal

CAFE: Cross-View Adaptive Fusion and Cluster Center Enhancement for Robust Multi-View Clustering.

IEEE transactions on pattern analysis and machine intelligence·2026

Same journal

DIVER: Reinforced Diffusion Breaks Imitation Bottlenecks in End-to-End Autonomous Driving.

IEEE transactions on pattern analysis and machine intelligence·2026

Same journal

Ethics-Aware Safe Reinforcement Learning for Rare-Event Risk Control in Interactive Urban Driving.

IEEE transactions on pattern analysis and machine intelligence·2026

Same journal

Learning Shape Anchors for Holistic Indoor Scene Understanding.

IEEE transactions on pattern analysis and machine intelligence·2026

See all related articles

TransFlow, a novel transformer architecture, enhances optical flow estimation by improving motion accuracy and recovering lost details. This method offers a simpler training approach and achieves state-of-the-art results on benchmark datasets.

Area of Science:

Computer Vision
Deep Learning
Artificial Intelligence

Background:

Optical flow estimation is crucial for computer vision tasks like motion estimation and object tracking.
Current Convolutional Neural Network (CNN)-based methods have limitations in accuracy and handling complex scenarios.

Purpose of the Study:

Introduce TransFlow, a transformer-based architecture for optical flow estimation.
Demonstrate TransFlow's advantages over existing CNN-based methods.
Evaluate TransFlow's performance on various computer vision benchmarks and downstream tasks.

Main Methods:

Utilized spatial self-attention and cross-attention mechanisms for enhanced correlation and matching.
Employed long-range temporal association to recover occluded or motion-blurred information.

Related Experiment Videos

Implemented a concise self-learning paradigm, removing the need for complex pre-training.

Main Results:

Achieved state-of-the-art performance on Sintel and KITTI-15 optical flow benchmarks.
Demonstrated superior performance in 3D scene flow estimation.
Showcased effectiveness in downstream tasks: video object detection, frame interpolation, and video stabilization.

Conclusions:

TransFlow offers improved accuracy and robustness in optical flow and scene flow estimation.
The transformer architecture effectively captures global dependencies and handles challenging visual data.
TransFlow serves as a versatile and effective baseline for future research in motion estimation.