Eager Few Shot OD with EfficientDet D2

[colabwarrior]

Prerequisites

Please answer the following questions for yourself before submitting an issue.

• [x ] I am using the latest TensorFlow Model Garden release and TensorFlow 2.
• [ x] I am reporting the issue to the correct repository. (Model Garden official or research directory)
• [ x] I checked to make sure that this issue has not already been filed.

1. The entire URL of the file you are using

https://github.com/tensorflow/models/blob/master/research/object_detection/colab_tutorials/eager_few_shot_od_training_tf2_colab.ipynb

2. Describe the bug

RetinaNet does a good job detecting the custom object when used in eager few shot learning code. I tried replacing RetinaNet with EfficientDet D2. Made the changes to shape i.e 768x768 where required. The model seems to be training but does not detect the custom object. I checked the checkpoint file and seems like no changes are required to load the model from checkpoint and also the "to_fine_tune" weights have the same name. Only thing that was required to be changed was the shape from 640x640 to 768x768 which I did. But EfficientDet D2 does not seem to detect anything.

Any help is appreciated on how to use EfficientDet models in eager few shot training mode.

6. System information

Running on google Colab Pro.

[ravikyram]

@colabwarrior

I have tried in TF nightly version. Please, find the gist here. You are also seeing the same behavior?
Thanks!

[colabwarrior]

Thanks for getting back on this. Yes, I do see the same behavior i.e. No bounding boxes around the object of interest when I use EfficientDet models.