Loss graph lost when training continues on colab

[vanquang203546]

For example, I trained 1000 batches. Then Colab disconnected and I retrained. The problem is that the previous Loss graph is lost. It only displays Loss from batches 1000 onwards
CLI: !./darknet detector train yolo.data cfg/v4mosaic.cfg backup/v4mosaic_last.weights -dont_show -map

[stephanecharette]

This is fixed on the new darknet/yolo repo: https://github.com/hank-ai/darknet

[Dickoabc123]

I am getting the problem similar to that, where I am only getting the mAP for every 1000 iterations, surely there must be a way to specify how many iterations to show the mAP. Would be better for every Epoch. Does anyone know how to show the mAP sooner than every 1000 epochs?

Id also like to be able to see the validation loss and the training loss.

Hi @stephanecharette , do you know how I could do this ?

[stephanecharette]

1. Darknet doesn't work with epochs. 2) You should use the Hank.ai repo instead. https://github.com/hank-ai/darknet#table-of-contents

[Dickoabc123]

@stephanecharette thank you for that. Does that give the validation losses ?
Since I already have my darker files set up and configured, is it easy enough to transfer across to Hank.ai ?