Train from the last checkpoint?

[RegisOlivo]

generated files: checkpoint, yolov3.index, yolov3.data-00000-of-00001

Or just start train.py again?

[Fafa-DL]

I have the same question,In Tensorflow1.x's version the Config's code are much different from version 2,How to deal with checkpoint?if I want to train from latest file?Thank you very much!

[moritz157]

I just changed the following in the train.py:

model = tf.keras.Model(input_tensor, output_tensors) #Already existing line
model.load_weights('./yolov3') #This is the new line

I don't know this for certain, but it seems like this just loads the saved weights and continues from there. I also see steady improvements using this method.

[MhdMartini]

I just changed the following in the train.py:

model = tf.keras.Model(input_tensor, output_tensors) #Already existing line
model.load_weights('./yolov3') #This is the new line

I don't know this for certain, but it seems like this just loads the saved weights and continues from there. I also see steady improvements using this method.

Did that line do the trick? I am running into the same issue now and just want to confirm.

[moritz157]

I just changed the following in the train.py:

model = tf.keras.Model(input_tensor, output_tensors) #Already existing line
model.load_weights('./yolov3') #This is the new line

I don't know this for certain, but it seems like this just loads the saved weights and continues from there. I also see steady improvements using this method.

Did that line do the trick? I am running into the same issue now and just want to confirm.

As far as I remember, it did.