search

sfzhang15 / RefineDet

Single-Shot Refinement Neural Network for Object Detection, CVPR, 2018
Other
1.43k stars 393 forks source link

anchor setting #165

Closed wait1988 closed 5 years ago

wait1988 commented 5 years ago

I noticed that each prediction layer has 3 anchors with the same aspect ratio,so is it enough for the task like person detection with various poses?And,how you choose the min_size and aspect ratio for each feature layer?

sfzhang15 commented 5 years ago

@wait1988 Each prediction layer has 3 anchors with one scale and three aspect ratios. If you want to detect persons, you can change the anchor setting according to your dataset. In general, we choose the min_size as 2~4 times of the stride size of the prediction layer. As for aspect ratio, we choose it according to the dataset.

wait1988 commented 5 years ago

@sfzhang15 Got it.You're so nice.