Source code using qqweee https://github.com/qqwweee/keras-yolo3, use your own do test data sets (6000 images), loss in the training to 11 no longer fell, in the compile and the statement of the display accuracy, has been to zero, parameters are preset, excuse me, how to solve?

CodePudding user response:

The original poster can use manual test data sets can be detected, parameter preset is what the devil, vector and see if it is too big

CodePudding user response:

refer to the original poster qq_31627815 response:

is qqweee https://github.com/qqwweee/keras-yolo3 source code, using its own do test data sets (6000 images), the loss to the 11 no longer fell during the training, in the compile shows accuracy of the statements, but have been zero, parameters are preset, excuse me, how should solve?

What's the matter with your training so fast? 272 ms/step

CodePudding user response:

Of the decline of the brothers you this loss is not solved, please

CodePudding user response:

Vector change loss function, adjust the relevant parameters and try again, no big problem

CodePudding user response:

The original poster hello, the situation I met with you, may I have your acc has been 0 problem solved? Thank you very much!

CodePudding user response:

Feeling is the issue of data sets

CodePudding user response:

Bloggers, want to consult the training accuracy when suddenly drop what reason be