I’m losing my mind here. I’ve been working on this stupid object detection model for weeks, and now that I’m finally training it, something’s clearly messed up. The training loop is running, epochs are ticking by, but every single loss value is a big fat ZERO. What gives? The code below me is an implementation of the EAST model and the reset 50 to help me out to localize my training data. Supposedly, I indented to be a simple version of it, expecting 4 bndboxes, for ymin, ymin, ymax, ymax, and the predicted class, and tweak it so it seems like and works for my old dataset loader that i used for my ssd training practice.