CVPR 2025 Best Paper Nominees

[Joanne]

CVPR 2025 Best Paper Nominees: Innovations Across Vision and AI

The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2025, one of the leading international conferences in computer vision, will take place from June 11 to June 15, 2025, at the Music City Center in Nashville, Tennessee.

This year, CVPR received a record-breaking 13,008 submissions, marking a 13% increase compared to last year, with 2,878 papers accepted at an acceptance rate of 22.1%. Among these, 16 exceptional papers have been nominated for the prestigious Best Paper Award, showcasing groundbreaking research across key areas:

3D Reconstruction & Rendering

• VGGT: Visual Geometry Grounded Transformer – Unified 3D reconstruction from various viewpoints.
• 3D Student Splatting and Scooping – Enhanced neural rendering methods for novel views.
• Difix3D+ – Diffusion-based refinement for improved 3D reconstructions.

SLAM & Scene Understanding

• MegaSaM – Robust SLAM solutions for dynamic scenes.
• Zero-Shot Monocular Scene Flow Estimation – Generalized scene flow estimation without explicit adaptation.
• GlobustVP – Robust vanishing point estimation for structured environments.

Depth Estimation & Stereo Matching

• FoundationStereo – Zero-shot stereo matching for robust depth estimation.
• TacoDepth – Efficient radar-camera fusion for precise, real-time depth estimation.

Vision-Language & Multimodal Models

• Molmo and PixMo – Open-source vision-language models competing with proprietary methods.
• Generative Multimodal Pretraining – Integration of language modeling and diffusion-based visual representations.

Diffusion Models & Generative Techniques

• Navigation World Models – Diffusion-based models for robotic navigation and planning.
• Reconstruction vs. Generation – Optimization strategies for latent diffusion models.

Image Editing & Processing

• AnyEdit – High-quality instruction-driven image editing platform.

Distributed & Hardware-Efficient AI

• Descriptor-In-Pixel – High-speed, on-chip feature tracking.
• UniAP – Optimizing parallelism in distributed deep learning systems.

Wildlife & Behavior Recognition

• PanAf-FGBG Dataset – Dataset highlighting background influences on wildlife behavior recognition.

Explore the nominated papers in detail on the official CVPR 2025 Award Candidates page.