Oddbean new post about | logout
 📝 Do We Still Need Non-Maximum Suppression? Accurate Confidence Estimates and Implicit Duplication Modeling with IoU-Aware Calibration 🔭

"Uses IoU-Aware calibration, a conditional version of Beta calibration to implicitly account for the likelihood of each detection being a duplicate and adjust the confidence score accordingly." [gal30b+] 🤖 #CV

⚙️ https://github.com/Blueblue4/IoU-AwareCalibration}{code
🔗 https://arxiv.org/abs/2309.03110v1 #arxiv

https://creative.ai/system/media_attachments/files/111/027/964/514/682/287/original/ee530fc0f8fdbd31.jpg

https://creative.ai/system/media_attachments/files/111/027/964/568/678/528/original/7f98891df7768e81.jpg

https://creative.ai/system/media_attachments/files/111/027/964/621/619/803/original/55ed4be077ec505a.jpg

https://creative.ai/system/media_attachments/files/111/027/964/693/493/377/original/84c3e51bde07eb57.jpg