I would like to create my own script to calculate precision and recall, how would it be possible for me to get the raw inference data after running tlt - evaluate or some other method. Then I can compute the recall, precision, myself.
How would I go about doing this for the SSD network for example?