Issue with Exercise 7 in C3_W2 Assignment

Course Q&A TensorFlow: Advanced Techniques Specialization Advanced Computer Vision with TensorFlow
1

At the end of Exercise 7, the test code requires the model weight shapes to be:

(3, 3, 256, 24)
(512,)
(256,)

I instead get:
(3, 3, 256, 24)
(256,)
(256,)

I create my box_predictor checkpoint with the _base_tower_layers_for_heads and _box_prediction_head loaded from my model. I then create the model checkpoint with this _box_predictor checkpoint and with the _feature_extractor from the model.

This is almost cut and paste from the interactive_eager_few_shot_od_training colab notebook.

Despite this I always get a value error regarding incompatible tensor shapes when running the prediction with my model. This only happens the first time, the second time I run it, it runs fine but then I get the wrong model shapes at the end as discussed.

My training loss doesn’t decrease and I’ve traced it back to this.

Any help would be really appreciated, I’ve been stuck on this assignment for about a week now :sweat:

2

This output obviously has to do with the layer heads you are choosing. I would have checked the code but I am only on my phone these days. Would you consider making a copy of this assignment for your records, reseting it and solving it again from scratch.

3

hi please i am facing the same problem , what is the cause ? thanks

4

Hi @Maya_Cerine_HAMMAMOU , @Joseph_Cameron,
Can you personally DM me your notebook. I will check and let you know.

5

Hope the solution fits. @Joseph_Cameron check exercise 6.2 there is a high possibility to be a mistake. You can follow TensorFlow 2 Object detection API Tutorial.
Regards
Arif

1 Like