about the model output number about nudenet HOT 4 CLOSED

Yes, that is correct. sexy + nude is unsafe and safe is safe.

from nudenet.

thanks for your reply. i train model with three class. use the seresnet50 as the backbone and finetune all layer with the imagenet model file. the loss is a little strange. Can you provide the train log about the acc and loss if you want.
this is my train log. I set batch size to 128.

LR:0.001 Training: Loss: 0.2197, Accuracy: 0.9197, Validation : Loss : 0.4271,  Accuracy: 0.8646, Time: 2284.8669s
LR:0.001 Training: Loss: 0.1424, Accuracy: 0.9486, Validation : Loss : 0.3773,  Accuracy: 0.8840, Time: 2300.8871s
LR:0.001 Training: Loss: 0.1224, Accuracy: 0.9558, Validation : Loss : 0.3751,  Accuracy: 0.8866, Time: 2290.4965s
LR:0.001 Training: Loss: 0.1100, Accuracy: 0.9602, Validation : Loss : 0.4759,  Accuracy: 0.8667, Time: 2310.8288s

from nudenet.

Can you provide the train log about the acc

I don't have the train log saved for the classifier. Sorry!

I haven't used seresnet before. Not sure how it is supposed to progress.

from nudenet.

thanks for your reply. give time a little time. the loss maybe go down :p

from nudenet.