I can't figure out how I can evaluate error metrics for the model I have already trained, taking advantage of protbuf and the tensorflow API.
The main problem is that the model is saved as .pb but in the zipper file I download I also have assets, variable. So to test it for example, I upload the zipper to colab and unzip it.
Then I use an infernce.py , exploiting the labelmap.pbtxt and giving the model path of the unzipped folder.
!python /content/inference.py -m /content/content/inference_graph/saved_model -l /content/testlabel_map.pbtxt -i /content/models/research/object_detection/test_images/my_test_folder
How do I calculate precision, recall, mAP? Could I also do this with Tensorboard? If so how?
I usually use the model.evaluate function, this is the first time I have done such work using protobuf, TFRrecords and tensorflow. I have no idea how I can manage to calculate the error metrics of my model.