install

pip3 install -r requirements.txt --user

launch server

NOTE! Ensure connections/ folder is deleted before running the app!

gunicorn --worker-class eventlet -w 1 app:app

Managing the collected data

To start from scratch, delete the games/ folder.

Then run cd data/; python3 generate_games.py

Throughout the data collection process, games will be marked as finished in the flags.json file.

Run this command to move the finished games to a folder called finished_games/: cd data/; python3 move_finished_games.py

Then, you will see how many games are left to be finished.

If you want to recollect those same games, run: generate_games.py and it will regenerate non finished games.

If you finish a round of data collection and there are still some games that haven't been finished, you can either leave them their for people to connect to automatically to finish the data collection, or you can delete them and generate all the games again.

Move the finished_games/ folder into the etc/DataViewer/static/ folder and then run run.py to visualize the collected data in your browser.

You can also move it into etc/estimate_dialogs_left_to_collect/ and run count_games.py to see how many conversations you have for each target image.

Creating target images

1. Batch download your images
2. Put them in a folder with the scale_crop.py script and scale them and crop them.
3. Run rename.py in the same folder as them (with optional argument to specificy what number you start at)
4. Generate the semantic labels (for me specifically, I put them in a folder called landscape/ on my server and run a command).
5. Organize your data so you have data/landscape_target/ where all your target images go with the extension .jpg. Have another folder called data/landscape_label/ where all your target label images go with the extension .png. NOTE: your target images and labels that correspond to each other must have the same root name. They only differ in the golder location and the .jpg and .png extension.

Coming soon: Generating data to input into the model from finished_games/ folder

Coming soon.