// the find
SkalskiP/top-cvpr-2025-papers
About This repository is a curated collection of the most exciting and influential CVPR 2025 papers. 🔥 [Paper + Code + Demo]
A yearly-refreshed list of CVPR papers (2023 through 2026 editions exist as siblings) with links to the arxiv paper, official code repo, demo, and video for each. It's for computer vision researchers and engineers trying to find the handful of papers from a 12,000+ submission conference that are actually worth reading, not a tool or library.
Every entry links straight to the GitHub implementation and often a live HuggingFace demo or Colab notebook, so you can go from title to running code in one click instead of hunting through a paper's appendix for a repo link. The list is generated from automation/data.csv via automation/generate.py rather than hand-maintained markdown, so updates are a data-entry problem, not a formatting one. Organization by topic (3D vision, pose estimation, VLMs, etc.) makes it skimmable instead of a flat 900-paper wall of text.
There's no selection criteria documented anywhere — 'most exciting and influential' is the author's personal taste with zero methodology, so you're trusting one person's read of a 2,878-paper accepted list. The repo is a point-in-time snapshot: once CVPR 2025 ends, this stops getting updates except for typo fixes, and the 🔥 highlight flags aren't explained (impact? personal favorite? something else?). It's not code you integrate or depend on — there's no package, no API, nothing to pip install, so its only use is as a reading list, and that value decays every year a new edition supersedes it.