search

Borda / keras-yolo3

A Keras implementation of YOLOv3 (Tensorflow backend) a successor of qqwweee/keras-yolo3
MIT License
31 stars 10 forks source link

porting to TF 2.0 #21

Closed Borda closed 4 years ago

Borda commented 5 years ago

Since this Keras implementation is based on TensorFlow, it makes sense to port ti to tf.keras in TensorFlow 2.0 or higher. This move may solve the issue with multi-GPU training #20

https://www.pyimagesearch.com/2019/10/21/keras-vs-tf-keras-whats-the-difference-in-tensorflow-2-0/

robisen1 commented 4 years ago

I would be interested in helping. That being said does the code not work with TF 1+?

Borda commented 4 years ago

well, in fact, it was already done :rabbit:

robisen1 commented 4 years ago

Sorry I do not understand. Are you saying you already have assistance? Are you staying that the code does work with TF1 and TF2? Thank you for responding.

Borda commented 4 years ago

it shall work now with TF 2.x, can you pls test it?

robisen1 commented 4 years ago

Yes! I will attempt to tonight or early tomorrow. That being said there is now no support for TF 1 right? Just want to make sure.

Borda commented 4 years ago

well not sure if it is back TF 1 compatible... :/

robisen1 commented 4 years ago

Well then i can test both. Ill make two different conda environments and report back.

robisen1 commented 4 years ago

Sorry I have no responded. I attempted to run the code but tensorflow but very upset due to the fact I have running cuda 11

In the meantime I am creating new environments with cude 10.1 so that everything will work.

Also.. this code base is very nice.

robisen1 commented 4 years ago

well not sure if it is back TF 1 compatible... :/

Would you mind emailing me. I would like to share some initial results from testing.

Borda commented 4 years ago

@robisen1 I think that the best way would be to share it here... =)

robisen1 commented 4 years ago

Ok I will later today. Some of it will be in the form of video showing inference.