VCode: a Multimodal Coding Benchmark with SVG as Symbolic Visual Representation

Kevin Qinghong Lin 1*, Yuhao Zheng 2*, Hangyu Ran 3*, Dantong Zhu 3, Dongxing Mao 3, Linjie Li 4, Philip Torr 1, Alex Jinpeng Wang 3✉
1University of Oxford   2University of Science and Technology of China   3Central South University   4Microsoft Research
* Equal contribution ✉ Corresponding author

Project Page Hugging Face Space Demo Code GitHub arXiv Paper Hugging Face Paper Page

🏆 Leaderboard

Loading Chart...
# Model Name VCode Score ↓ General Professional Vision-centric SigLip Score Code Token (K) Success Rate