EfficientDet TFrecords size issue

size of the tfrecords generated by efficientdet for training and validation dataset is much higher than dataset itself.

Did you follow jupyter notebook? Can you reproduce with it?

yeah. I have followed that. and trained efficientdet on custom dataset. tfrecords are generated fine. but it takes 35GB storage for 56k images.

How about using the coco dataset mentioned in the notebook?

it took 20GB

So, how about the size of coco dataset and also your 56k images?

for coco dataset size is ~20GB and for my dataset the size is ~35GB.
My concern is Tfrecords size in efficientdet is much higher when compared with faster_rcnn. when faster_rcnn tfrecords are created with the same dataset its size is in MB

You can inspect the feature inside the tfrecord files. It contains lots of features.
It is not the same as faster_rcnn.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.