Training SSD with TFRecords in TLT

I am using the latest TLT container nvcr.io/nvidia/tlt-streamanalytics:v3.0-dp-py3

I used the sample jupyter notebook provided in TLT container to train ssd, when I changed the data source to tf_records instead if images files
data_sources {
tfrecords_path: “/workspace/tlt-experiments/data/tfrecords/kitti_trainval/*”
image_directory_path: “/workspace/tlt-experiments/data/training”
}
it gives an error
google.protobuf.text_format.ParseError: 48:5 : Message type “DataSource” has no field named “tfrecords_path”.

However, training with tf records works fine with detectnet and FRCNN, the config parser accepts tfrecords_path without any problem!

See SSD — Transfer Learning Toolkit 3.0 documentation
In SSD, please set images or labels path directly.

data_sources: {
label_directory_path: “/path/to/train/labels”
image_directory_path: “/path/to/train/images”
}

Thanks for you answer.
I understand if we used it as it is it would work.
I would like to use TFRecord instead of reading the images from the files, for faster training as same as detectnet and FRCNN

I will sync with internal team for your request.