search

weiliu89 / caffe

Caffe: a fast open framework for deep learning.
http://caffe.berkeleyvision.org/
Other
4.77k stars 1.68k forks source link

person detect : finetune with VGG_coco_SSD_300x300_iter_240000.caffemodel get higher recall ,but higher false positive #229

Open darrenlei opened 8 years ago

darrenlei commented 8 years ago

i want to do person detection, and finetune with VGG_coco_SSD_300x300_iter_240000.caffemodel, and get higher recall, but also higher false positive.

person detect :the original model detect result : Recall:50.65%, fault:9.97%, the finetune iteration 5000 result: Recall:59.80%, fault:24.12%

any one could help me , to improve recall ratio, and decrease false positive ratio ?

weiliu89 commented 8 years ago

Was the fault caused by poorer localization or mis-classification?

dingdashen commented 7 years ago

can you tell me how to get the recall? Thank you