Program was created by Daniel Persson and Petter Gullin. This program was created for our B-essay (B-uppsats), the B-essay is included in the repository.
Thank you Albin Larsson (Abbe98 on github) for all the techincal help and guidance on this project
| Programfolders | Start files |
|---|---|
| trainingmaterial_app | main.py |
| Trainingtensorflow_scripts | create_training_data.py, create_training_data.py |
| Validation_scripts | compare_folders.py, manual_validation.py, validation.py |
These programs download, creates and validates picture from Swedish National Heritage Board (Riksantikvarieämbetet) database. This throught API ksamsok.
These programs were created as the Swedish National Heritage Board saw a problem with the quailty of the metadata.
To find wrongfully classifed metadata in their database.
Here comes lists about the programs.
- The program either validates from a retrained model or a custom trained model.
- The files in this repository is the two best models from our testing trained on aprox 22000 pictures.
- The image csv files are before the balancing of the two folders to equal levels.
- Handels the downloading of the pictures
- Deleting all corrupted files
- Removing the info of deleted files from the csv files.
- Randomising image files in a folder
- Creates the trainingdata
- For the retrained model use this tutorial
- Validate against the database by instution with the wrongfully validated pictures getting downloaded
- Validate manually from either a folder or a file
- Compare two folders and printout the number of same files in the folders