Purpose: 😄

Folder 📂	Description
Face Detection	Face Detection folder
TF-retrain	Retrain google classifier

This is annoying to have multiple pictures from multiple camera, sorted all a different way.
It would be simpler to sort photos by person, place, date, camera, landscape, portrait, etc ...
Photos.app from  already does all of that but i wanted to use Machine Learning to try it myself with python.

First step:

Face Detection

On the web some tutorials are very poor and useless, however i found this one https://realpython.com/face-recognition-with-python/#want-to-know-more.

Thank you - Shantnu Tiwari - ❤️

This one is very clear, short and well-written. The programme use the famous open librarie OpenCV.

OpenCV uses machine learning algorithms to search for faces within a picture. OpenCV use cascade classification and more precisely Haar Feature-based Cascade Classifier for Object Detection. Therefore the Classifier is train with tons of data and pretty quickly, resulting in a XML file. Since face detection is such a common case, OpenCV comes with a number of built-in cascades for detecting everything from faces to eyes to hands to legs.

You can use the first programme as follow:

python3 face_detect.py myPhoto.png haarcascade_frontalface_default.xml

The result are good with a high detection rate. However time to time, i will or will not detect face. To get around this problem, you can adjust the scaleFactor parameter.

This face detection option is native in numerous of devices as camera, computer, phone, etc… My next goal is to detect and recognise.

Second step

Recognise

https://www.tensorflow.org/hub/tutorials/image_retraining

Tensorflow is an open source machine learning framework developped by Google. The one resource, i will use is How to Retrain an Image Classifier for New Categories in the TF Hub section.

First of:

pip install tensorflow-hub

Then the you need some photos of whatever you need to be recognise. In my case, i chose Harry Potter, Hermione, Ron from the Harry Potter saga. To download all pictures (~100 per personages), i used a script named googleimagesdownload. This was the best way i found to get photos, api for Bing or Google was to laborious for this simple purpose.

Once you have all pictures, make sure to order the different folder correctly.

ex: ~/downloaded | |-Harrypotter ~100 |-Hermione ~100 |-Ron ~100

Some pictures downloaded was awful so sort out bad picture or unrelated photos.

curl -LO https://github.com/tensorflow/hub/raw/master/examples/image_retraining/retrain.py

Or you can find the script in the TF-retrain/retrain.py

Run python retrain.py -h for all options.

However in a simple way you can run: ~20 minutes

python retrain.py --image_dir ~/downloaded

The script will write out the new model trained on your categories to /tmp/output_graph.pb, and a text file containing the labels to /tmp/output_labels.txt

Now the model is train to recognise and sort a picture between the 3 folders (or numbers of folder you have).

We need to tell the classifier to use our model:

curl -LO https://github.com/tensorflow/tensorflow/raw/master/tensorflow/examples/label_image/label_image.py
python label_image.py \
--graph=/tmp/output_graph.pb --labels=/tmp/output_labels.txt \
--input_layer=Placeholder \
--output_layer=final_result \
--image=IMAGE_TO_SORT

For some details, refer to this video from channel “google developers”: https://www.youtube.com/watch?v=cSKfRcEDGUs

I do recommand you to read this article coding-robin.de/train-your-own-opencv-haar-classifier which use a different approach.