IDEA-Research / GroundingDINO

[ECCV 2024] Official implementation of the paper "Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection"
https://arxiv.org/abs/2303.05499
Apache License 2.0
6.4k stars 661 forks source link

Why is the effect better on the web page? #317

Open pzhren opened 5 months ago

pzhren commented 5 months ago

https://deepdataspace.com/playground/grounded_sam。Why can you get better results in this link, but worse in github? image image BOX_THRESHOLD = 0.35; TEXT_THRESHOLD = 0.25

songrise commented 3 months ago

Hi, I also found similar issue. I guess it might due to 1. They use bigger model on web. 2. they use different threshould. Do you figure out how to improve the huggingface model?