Custom dataset

[upjabir]

can you please suggest how to detect custom image dataset?

[IdeaKing]

Hi @upjabir I've been working on creating custom YOLOv3 too. I tried this repo but for me, it doesn't work. I only get one bounding box in the output image.

I suggest that you look around for other YOLOv3 implementations because you will need to write more functions in this repo to detect your own classes

[RahmadSadli]

@IdeaKing This code works perfectly as mentioned in the tutorial using Pyhton 3.7 and Tensorflow 2.0, (Not yet tested on TF 2.1 or above, I will update it in the near future)

It 's possible you'll get the following error if you just installed your TensorFlow 2.0,
This is usually caused by the broken TensorFlow-estimator module that's could influenced by the higher TensorFlow-estimator module.

AttributeError: module 'tensorflow' has no attribute 'compat'

What you need to do is simply run this command:

pip install tensorflow-estimator==2.0.*

and it works.

Here is an example of the result:

@upjabir As this code is targeted for beginners to help them in understanding the basic concept of how Yolov3 works, it only can be used with the pretrained Yolov3's official weights. For the custom dataset, there are still a lot of works to do including calculating the losses (GIoU or DIoU that depends on which one you want to use) to train the network. Good luck !

[RahmadSadli]

@IdeaKing just tested on this environment:

TF 2.1
Python 3.7
and OpenCV 4.4.0.
(I'm using Anaconda on Windows 10)

It works, I found no problem.

I have no idea why in your case only get one bounding box in the output image. can you please give a detailed of your working environment?

[IdeaKing]

Hi @RahmadSadli, I think it was because I was running on Google Colab and I may have messed up somewhere. I retried your tutorial and it works now. Thank you for your help, and sorry for any inconvenience.

[RahmadSadli]

Hi, @IdeaKing, glad to hear that..no problem..

[robbinspannier]

@IdeaKing just tested on this environment:

TF 2.1
Python 3.7
and OpenCV 4.4.0.
(I'm using Anaconda on Windows 10)

It works, I found no problem.

I have no idea why in your case only get one bounding box in the output image. can you please give a detailed of your working environment?

Hello,
first of all: Thank You for all the effort You put into this project. It was really helpful.

I had the same issue like IdeaKing. I only got one bounding box in the output images as well. I could only fix this by just using your complete code directory from github. ---> So there might be a small error in one of your 4 tutorials that causes the issue.