IDEA-Research / DINO

[ICLR 2023] Official implementation of the paper "DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection"
Apache License 2.0
2.17k stars 235 forks source link

How to label negative samples (without any object)? #262

Open brent88310070 opened 3 months ago

brent88310070 commented 3 months ago

Hi everyone,

I have two questions about negative samples, which are defined as images without any objects. My task involves many negative samples, and I need to determine whether the output is negative or not.

Question 1: Despite related issues before ( #38 ), I still can't figure out how DINO classifies negative samples.

Question 2: If DINO can classify negative samples, how should these samples be labeled ? Should all values be zero (e.g., label = 0, bbox = [0,0,0,0]) ?

Thanks a lot!

RunkunL commented 1 week ago

I have the same problem, I wonder if you have solved it. If possible, can we discuss it?