Spotted!

Acknowledgements

Thanks to pyimagasearch.com and many others...

What is this?

Spot is a tool capable of selecting interesting pictures within a given dataset. It uses the Structural Similarity Index (SSIM) to quantify the differences between two images. It is also capable of performing image classification using a Support Vector Machine (SVM). Additionally, it provides a script to select sunset/sunrise/dawn/dusk images looking at the image filename, e.g. 1529032200_0_2018-06-15-04-10-00.jpg.

Dependencies

It utilises: scipy scikit-image opencv-python sklearn numpy astral pytz datetime

How to use

Create a Python3 virtual environment following this. Within the virtual environment pip install -r requirements.txt

In the COMMON section of the conf file, specify EXTENSION of the image to be processed.

SSIM SELECTION:

In the SSIM section of the conf file:

• Specify PATH_IN, PATH_OUT, REF_IMG_PATH, and THRESHOLD.

Run source select.sh <path/to/config-file> <mode> where <mode> has to be set to "quick". The other mode is "full", and it runs the timestamp selector first, then the ssim.

TIMESTAMP SELECTION:

The tool is designed to read images whose file names are date stings with a format specified in the conf file

DATE_FMT = %Y-%m-%d-%H-%M-%S

Change this at your own risk. The code hasn't been run with any other format.

In the TIMESTAMP section of the conf file:

• Specify PATH_IN, PATH_OUT, TIMEZONE, and CITY_NAME.

Run within repo main dir

python src/timestamp-filter.py -c <path/to/config-file>

SVM

Training

In the COMMON section of the conf file:

• specify SVM_PATH where the trained svm will be saved
• specify the number of clusters N_CLUSTERS

In the TRAINING section of the conf file, specify TRAINING_PATH.

Run source trainSVM.sh <path/to/config-file>

Classification

In the CLASSIFICATION section of the conf file:

• specify TEST_DIR (directory with images to be classified)
• specify JSON_FILENAME (file with results)

Run source classify.sh <path/to/config-file>