Object Detection

A small project utilizing RetinaNet object detection module developed by Tsung-Yi Lin at Facebook AI Research (FAIR), implemented with Keras, (the original project by FAIR is called Detectron). For the paper and libraries used, see References below. For installation, see instructions

Steps taken inside the code

1. Declare object detection module as implemented by AI Commons
2. Set RetinaNet as the model type.
3. Use a pre-trained model. You can find the latest version here or on Kaggle (search yourself.)
4. Magic

That's it.

Installing

pip install tensorflow numpy scipy pillow matplotlib h5py keras opencv-python

Running

Default parameters:

• An image for the input, you can put it on the same directory. Default name for image is ilia.jpg
• Default model: resnet50_coco_best_v2.1.0.h5.

python objD.py

If everything runs smooth as criminal, you should see iliaRevealed.jpg

Did you make the whole thing?

No. As said, I am only using algorithms and pre-trained models that were developed by FAIR and ImageAI. The reason why this is on GitHub is because later on I'm planning to do use it for future projects. The reason why it's public is because of Contributor Agreements, so this project, despite having 30 lines of code, has to be open-sourced.

Trouble? Questions?

Submit an issue.

Training your own model

Good luck with that

References

• The paper
• AI Commons
• Detectron
• Keras RetinaNet
• ImageAI
• TensorFlow
• Keras