How to make pre-trained models

[hyejun]

Hello.

It is good work. I ran bonnet docker and it works well. I download pre-trained models (people) and tested. https://www.youtube.com/watch?v=rl0X4XKp4k0

Actually I am beginner on tensorflow/CNN. If I want to make a pre-trained model, What should I do?

How to use existed dataset and model for making the pre-trained model?

• for example, I want to use coco dataset.
• I found dataset on http://cocodataset.org/#download, and how to make pre-trained model (data.yaml, net.yaml, train.yaml)
• Can I select some objects for segmentation?
• I found some frozen protobuf, can I use this right now? https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md

Actually I finally want to make pre-trained models related to indoor environment (wall, door, ground) for robot navigation.

I appreciated your work and I agree what you said.

We strongly believe that our framework allows the scientific robotics community to save time on the CNN implementations, enabling researchers to spend more time to focus on how such approaches can aid robot perception, localization, mapping, path planning, obstacle avoidance, manipulation, safe navigation, etc.

Thank you.