search

msracver / FCIS

Fully Convolutional Instance-aware Semantic Segmentation
MIT License
1.57k stars 414 forks source link

If my data is small,such as 6000 images,is it difficult to train the model? #144

Closed CPFelix closed 5 years ago

CPFelix commented 5 years ago

I got my loss at: 2018-12-05 09:12:04,200 Epoch[9] Train-RPNAcc=0.929830 2018-12-05 09:12:04,200 Epoch[9] Train-RPNLogLoss=0.219037 2018-12-05 09:12:04,200 Epoch[9] Train-RPNL1Loss=0.138239 2018-12-05 09:12:04,200 Epoch[9] Train-FCISAcc=0.954637 2018-12-05 09:12:04,200 Epoch[9] Train-FCISAccFG=0.215569 2018-12-05 09:12:04,200 Epoch[9] Train-FCISLogLoss=0.174942 2018-12-05 09:12:04,200 Epoch[9] Train-FCISL1Loss=0.012621 2018-12-05 09:12:04,201 Epoch[9] Train-FCISMaskLoss=0.374442 2018-12-05 09:12:04,201 Epoch[9] Time cost=4398.860 And its mAP is all 0.How it happened? Anyone meet this problem?

CPFelix commented 5 years ago

If you use your own data,please be sure your code to get the data is right.

ccxietao commented 5 years ago

I got my loss at: 2018-12-05 09:12:04,200 Epoch[9] Train-RPNAcc=0.929830 2018-12-05 09:12:04,200 Epoch[9] Train-RPNLogLoss=0.219037 2018-12-05 09:12:04,200 Epoch[9] Train-RPNL1Loss=0.138239 2018-12-05 09:12:04,200 Epoch[9] Train-FCISAcc=0.954637 2018-12-05 09:12:04,200 Epoch[9] Train-FCISAccFG=0.215569 2018-12-05 09:12:04,200 Epoch[9] Train-FCISLogLoss=0.174942 2018-12-05 09:12:04,200 Epoch[9] Train-FCISL1Loss=0.012621 2018-12-05 09:12:04,201 Epoch[9] Train-FCISMaskLoss=0.374442 2018-12-05 09:12:04,201 Epoch[9] Time cost=4398.860 And its mAP is all 0.How it happened? Anyone meet this problem?

Hello, I also trained my own data set, but map=0, how do you solve it

CPFelix commented 5 years ago

@ccxietao You should check your code for read your own data,it probably has some problems.