search

sanghoon / pva-faster-rcnn

Demo code for PVANet
https://arxiv.org/abs/1611.08588
Other
650 stars 241 forks source link

caffe loss=0, after relabeling and debugging for a million of times #103

Closed dh928 closed 3 years ago

dh928 commented 4 years ago

I was training my model using caffe. it worked for the first group of samples, but it just doesn't work for the second group. my mbox loss is always zero and the accuracy is zero too. I changed all the paths correctly. Since for the first group I used the newer version of rs label, and for the second group I used an older version. I relabel all the samples so they can match. when I retrain the second group it just wouldn't work