Zhedong Zheng
Zhedong Zheng
@windpls 1.I use mirroring the image and adding dropout 2. I test model on fddb and tune the threshold according to the recall rate in paper. after 12net 12netc 94%...
@windpls after 12net 12netc and nms it remains average about 1000(this number I don't remember clearly) after 24net 24netc and nms it remains average about 100(this number I don't remember...
@tangtangsiqi "_12" means code for matconvnet-1.12 which is not needed. You can delete them. About data, you can refer to https://github.com/layumi/2015_Face_Detection/issues/1
@argman My email is [email protected]
@chichan01 As the original paper said, on fddb we should enlarge bounding box. I enlarge it by mutiple height with 1.4. (It's really important! ) My advice is to y1-(y2-y1)*0.4....
@chichan01 I suppose that bounding box can be represented as (x1,y1)left-top point (x2,y2) right-down point To enlarge the bounding box, you may change y1 = y1-(y2-y1)*0.4
@chichan01 I suggest that you may visualise the bounding boxes on image. It can help you to choose a better enlarge factor. I get 0.4 height by using this method.
@chichan01 Yes, they're the latest model which I trained.
@chichan01 Yes, this code didn't get the same result. As far as I concerned, some false-costive sample is selected. So the curve didn't perform well at first 50.
@Excaliburyz 1.custom layer is to connect 12net fc layer to 24net. custom48 layer is to connect 24net fc layer to 48net. 2.I run code in Cpu mode, which cost 3-4...