I am not getting good results from this assignment. For reference, here is my learning curve
Start fine-tuning!
batch 0 of 100, loss=1.6413319
batch 10 of 100, loss=0.91598403
batch 20 of 100, loss=0.56793094
batch 30 of 100, loss=0.42211616
batch 40 of 100, loss=0.35207018
batch 50 of 100, loss=0.31909102
batch 60 of 100, loss=0.30835617
batch 70 of 100, loss=0.30452627
batch 80 of 100, loss=0.30302587
batch 90 of 100, loss=0.30234587
Done fine-tuning!
Here is my predictions
237
(4,)
False
False
False
I am not getting any bounding box to show up on the three images despite using option 2. And my score is 0.0.
I understand that there can be many potential mistakes. One part that I am not sure of is that, when doing exercise 6, I only used detection_model._box_predictor._base_tower_layers_for_heads and detection_model._box_predictor._box_prediction_head, and didn’t use detection_model._box_predictor._prediction_heads. In fact, for some reasons, detection_model._box_predictor._prediction_heads does not appear to be a dictionary with 2 keys as stated, instead, it appears as a dictionary with 1 key. I don’t know whether this is related, but any idea would be appreciated. Thank you!