Datasets:

---

bop-benchmark

/

bop_extra

like 1

Follow

BOP: Benchmark for 6D Object Pose Estimation 82

Hi, I provided the updated cnos25 2D detection and segmentation baseline. Given that it

• gives a 10% absolute and 30% relative mAP gain on the tested H3,
• is conceptually equivalent to the simple CNOS baseline, without adding complexity,
• is fast (Avg. 0.134 sec / image on RTX4090),
• won the "Best Fast Method" and "The Best Open Source Method" on Track 7 in 2025,

I wonder if the organizers have any interest in updating the default detections with the results from this method.
If so, I can gather and provide results for the other datasets as well. Not sure about your plans for next year, though, so please let me know.
Thanks.