Hello everyone! Firstime using Weights and biases. Came from 2 minute papers and was pleasantly surprised when I heard 3 blue 1 brown in the promo video also! Must be a good tool if 2 of my favorite YouTubers are involved!
Since I’m a new user I included all of my resources links and images in this paste bin:
the last link in the above paste is the link to the imgur images
I’m developing a Car Object Detection model for a university project using the TensorFlow 2 Object Detection API [1].
To get to the point I’m at now where I have a trained model for object detection I was following this tutorial: [2]
My question is how can I Integrate W&B into a "TensorFlow 2 Object Detection API " workflow?
I’ve searched around the internet and only found these 2 related questions:
[3]
[4] (first top comment is the same question)
Both of the above sources are unanswered. So I’m sure that some other people in the future might come across this same unanswered problem.
From what I understand W&B works based on callbacks from the model.fit function like so:
[6]
But the "TensorFlow 2 Object Detection API " doesn’t directly use the model.fit function but it calls a python script like so:
[7]
(For training I then paste this command into the terminal alternatively I could also just paste it into jupyter)
I asked chat GPT about this problem and this is the answer it provided me:
[8]
I tried Chat GPT’s solution but the results are a bit weird:
First successful run:
logs:
[9]
Charts:
[10]
My question is what is the correct way to integrate W&B into the "TensorFlow 2 Object Detection API " workflow? Have I done it correctly but am I missing something? Also, where do I specify that W&B should keep track of the loss and other variables or is that done automatically?
Here is what my training looks like in the terminal for more info:
Executing the training command:
[11]
…
A lot of skipped terminal lines
…
Epcoh result logging after the training kicks off:
[12]
I hope someone can help me out and thanks for all the help I advance!
