Welcome to this project! This is a fork from TensorFlow Model Garden (https://github.com/tensorflow/models), to simplify for those who are already familiar with that repository.
This project helps you to automatically create labels in your images for object detection problems training your model in a semi-supervised way. The process is done by some algorithms and modifications added in the original repository.
The quick explanation of how it works is here: https://towardsdatascience.com/tensorflow-semi-supervised-object-detection-architecture-757b9c88f270#1a19-7d5b5cba66d1
A most detailed one is here: https://www.researchgate.net/publication/343167804_Semi-supervised_approach_to_auto-label_images_in_object_detection
- Fork this repository
- Enter in the research/object_detection folder
- Replace the images and XML files in train_images and test_images folder by yours.
- Added all your unlabeled data into the unlabeled_data folder
- Push your changes
- Go to the Google Colab (https://drive.google.com/file/d/1srnHOXFv779IGT0lKKGbTn8J-5dGSAX_/view?usp=sharing) make a copy in your drive, change the repository in the first cell by yours and play the run button
- That's it.
If you want to help this idea to get better be free to open issues and collaborate. Have a good insight? Let's talk!