CODEWITHSUNDEEP
deep-learningdata-conversion
Nick Skywalker
Nick Skywalker

Reputation: 1087

Convert custom tiny-YOLOv3 to a tensorflow format

After having tried all solutions I have found on every github, I couldn't find a way to convert a customly trained YOLOv3 from darknet to a tensorflow format (keras, tensorflow, tflite)

By custom I mean:

So far I am happy with the results on darknet, but for my application I need TFlite and I can't find working method for conversion that suits my case.

Anyone have succeed in doing something similar?

Thank you.

Upvotes: 1

Views: 2788

Answers (2)

Tanveer
Tanveer

Reputation: 900

This is the most simplest and easy repo. Author has done a wonderful job and it works well with yolov3, yolv3-tiny and yolov-4. Please don't forget to change the coco.names under classes if you are training on custom classes.

Git link for the code

Upvotes: 0

hiranyajaya
hiranyajaya

Reputation: 569

Do you have the resulting .weights file for your custom model?

If so, the following project by peace195 may help: https://github.com/peace195/tensorflow-lite-YOLOv3

EDIT:

In the above link, use convert_weights_pb.py file to convert your .weights file to a .pb file.

Then use the .pb file as a saved model and convert it to a .tflite model using the following command.

tflite_convert --saved_model_dir=saved_model/ --output_file yolo_v3.tflite --saved_model_signature_key='predict'

Thanks Anton Menshov for your suggestion on improving the answer.

Upvotes: 3

Related Questions