search

andy-yun / pytorch-0.4-yolov3

Yet Another Implimentation of Pytroch 0.4.1 and YoloV3 on python3
MIT License
278 stars 72 forks source link

Can you tell me meaning of loss values? #75

Closed richard0326 closed 5 years ago

richard0326 commented 5 years ago

if Layer(106) nGT 2, nRC 1, nRC75 0, nPP 7, loss: box 1.454, conf 0.942, class 0.152, total 2.548

is printed... What does Layer, nGT, nRC, nRC75, nPP means?

andy-yun commented 5 years ago

@richard0326 nGT: number of ground truth nRC: number of recall > 0.5 nRC75: number of recall > 0.75 nPP: number of proposals

richard0326 commented 5 years ago

What does Layer mean in this training??

richard0326 commented 5 years ago

Layer ( n ) 괄호 안에 있는 n의 의미가 뭔가요?

andy-yun commented 5 years ago

@richard0326 Layer(n) means the nth layer in the yolo v3 structure. you can easily find the yolo layers in the structure. n means the position of [yolo] layer.

richard0326 commented 5 years ago

Thank a lot!

richard0326 commented 5 years ago

I have one more question

8: Layer(082) nGT 27, nRC 4, nRC75 2, nPP 0, loss: box 128.327, conf 8.406, class 19.071, total 155.805 8: Layer(094) nGT 27, nRC 11, nRC75 2, nPP 16224, loss: box 8.211, conf 1075.647, class 23.411, total 1107.269 8: Layer(106) nGT 27, nRC 0, nRC75 0, nPP 64896, loss: box 21.769, conf 5427.244, class 23.114, total 5472.127

in training I get 3 layer result for each step... Why do I have 3 output for one training??

is it just output of the middle layer?