ResNet and YOLO Win Time-Tested Awards at CVPR 2026 – Full Award Breakdown

CVPR 2026 was held in Denver, Colorado from June 3‑7. The conference received 16,092 paper submissions, of which 4,071 were accepted (25.3% acceptance rate), marking a 23.71% increase in submissions over the previous year.

The conference released extensive data analyses covering submission trends, reviewer and area‑chair counts, geographic distribution of authors and reviewers, and the computational resources used by the community (average 4 GPUs per researcher, 40 GB memory per project).

Best Paper Awards

Five papers were honored: one Best Paper, two Best Paper nominations, one Best Student Paper, and one Best Student Paper nomination.

Best Paper

Efficiently Reconstructing Dynamic Scenes One D4RT at a Time – DeepMind, UCL, Oxford (arXiv:2512.08924). The authors replace fragmented frame‑by‑frame decoding with an on‑demand query paradigm (D4RT). A global scene encoder compresses the video, and a lightweight decoder answers arbitrary 3D queries (depth, point cloud, trajectory, camera pose). The method sets new SOTA on dynamic 4D reconstruction and tracking, delivering faster, more accurate dense reconstructions.

Best Paper Nominations

SAM 3D: 3Dfy Anything in Images – Meta AI (arXiv:2511.16624). Introduces a generative model for visually‑grounded 3D object reconstruction from a single image, leveraging a human‑and‑model‑in‑the‑loop pipeline, synthetic pre‑training combined with real‑world alignment, and achieves at least a 5:1 win in human‑preference tests.

NitroGen: An Open Foundation Model for Generalist Gaming Agents – NVIDIA, Stanford, Caltech, University of Chicago, UT‑Austin (arXiv:2601.02427). Presents a vision‑action foundation model trained on >1,000 games and 40,000 h of gameplay video. Core contributions: a large‑scale video‑action dataset, a multi‑game benchmark, and a behavior‑cloning trained unified model, yielding up to a 52% relative boost in task success over from‑scratch baselines.

Best Student Paper

Native and Compact Structured Latents for 3D Generation – Tsinghua, Microsoft Research, USTC, Microsoft AI (CVPR virtual poster 37074). Proposes O‑Voxel, a sparse voxel structure that jointly encodes geometry and appearance, robust to arbitrary topology. A sparse‑compression VAE with 4 billion parameters achieves high compression and superior geometry/texture quality.

Best Student Paper Nomination

ChordEdit: One‑Step Low‑Energy Transport for Image Editing – Guangdong University of Technology, Huizhou University, Shenzhen University, Peking University (arXiv:2602.19083). Reformulates text‑guided image editing as a transport problem between source and target distributions, deriving a low‑energy control strategy from dynamic optimal transport theory. The model‑agnostic method enables fast, lightweight, high‑fidelity single‑step editing.

Longuet‑Higgins (Time‑Tested) Prize

Two seminal papers received this award:

Deep Residual Learning for Image Recognition (ResNet) – Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun (2015). Introduced residual connections to solve gradient degradation, becoming the backbone for modern CNNs, Transformers, and large‑language models; cited >320,000 times.

YOLO v1 – Joseph Redmon et al. (2016). Pioneered single‑stage, end‑to‑end object detection (You Only Look Once), achieving 45 FPS on Titan X (Fast YOLO 155 FPS) and catalyzing SSD, RetinaNet, and the YOLO family; cited ~80,000 times.

Youth Scholar Award

Recognized Deepak Pathak (Carnegie Mellon) and Vincent Sitzmann (MIT) for outstanding contributions across computer vision, machine learning, robotics, neural scene representations, and generative modeling.

Thomas S. Huang Memorial Award

Awarded to Noah Snavely (Cornell) for exemplary research, teaching, and service in computer vision and graphics.