The maximum number of predicted instances of blendmask network is 100

Open disciplinejzx opened this issue 3 years ago • 1 comments

Running result of demo/demo.py: [11/16 21:14:27 detectron2]: ../test_img/8/x2.tif: detected 100 instances in 3.65s

Hello, when the number of instances in the test picture I used exceeds 100, the result of running blendmask can only predict 100 instances, and the target of instances outside 100 instances cannot be predicted and segmented. However, when I cut the test picture until the number of instances is less than 100, all instances in the picture can be segmented by prediction.

How can I modify the code to increase the upper limit of the prediction instance?

Nov 16 '22 14:11 disciplinejzx

adet/config/defaults.py set _C.MODEL.FCOS.POST_NMS_TOPK_TEST = max number ofinstances

Oct 27 '23 09:10 xunzha