something about 'MAX_NUM_GT_BOXES'

ruoqianguo / DetNet_pytorch

An implementation of DetNet: A Backbone network for Object Detection.

MIT License

377 stars 117 forks source link

something about 'MAX_NUM_GT_BOXES' #3

Closed baolinhu closed 6 years ago

baolinhu commented 6 years ago

I want to use your code to train your own dataset. There are many gt boxes in it, and there are probably hundreds of them.MAX_NUM_GT_BOXES=20, can I change this parameter a lot? I see the default in Pascal VOC is 20.

ruoqianguo commented 6 years ago

i think you can change that if you need (if your data gt boxes are less than 20).

baolinhu commented 6 years ago

@guoruoqian Thanks for your reply. I mean my dataset has much more objects in one image.

1.If Isets MAX_NUM_GT_BOXES=20,This means that if my GT boxes is greater than 20, will there be some boxes ignored?
2.I want to set MAX_NUM_GT_BOXES=100.Is there some bad effect? It will train slower or effect AP?

ruoqianguo commented 6 years ago

@baolinhu ,hi

if your GT boxes is greater than 20, there will be some boxes ignored.
I think if you set MAX_NUM_GT_BOXES=100. it will train slower and not effect AP.