I've been studying and using the Tensorflow Object Detection API for a couple of weeks. I've successfully trained a model. Now that it's ready, I'm diving into the details about the training and something that I couldn't find an answer is:
Which kind of evaluation metrics was used during the train?
I'm using the ssd_mobilenet_v1_coco_11_06_2017 pre-trained model and in the config file, I have:
eval_config : {
num_examples : 40
}
I thought initially that I was using some kind of logarithmic loss metric but searching in-depth in the repo, I couldn't find anything helpful that prove what kind of metric I'm using, neither what was the loss function used. The training output looks like this:
Hope you could help me with this! :)
Researching more about metrics and reading more here. I've finally found out that I wasn't using any metrics during the training. Those kind of metrics would be used only if I had something like:
eval_config : {
num_examples : 40,
metrics_set: 'pascal_voc_detection_metrics'
}
The documentation about available metrics to be used are available here: