Exporting object detection inference graph using custom input dimensions, with the option to remove preprocessing steps

Prerequisites

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

• [x ] I checked to make sure that this feature has not been requested already.

1. The entire URL of the file you are using

https://github.com/tensorflow/models/tree/master/research/object_detection/...

2. Describe the feature you request

Current implementation of TensorFlow 2 object detection api allows for exporting of a trained models inference graph using only an image tensor, a float image tensor, an encoded image string tensor and tf example string tensor. It would be extremely useful to be able to define custom input tensor dimensions i.e. something which has more than 3 channels, as well as have the option to remove any preprocessing steps such as image resizing. (My custom trained model never encounters any different input dimensions or data types besides float32).

3. Additional context

If anyone has any advice on how to add these features please let me know! And also if there already is an option/method to remove any preprocessing steps when exporting a model please forgive my ignorance. Thank you!!

4. Are you willing to contribute it? (Yes or No)

Yes!!