Read the Blog about this project.
Check the demo hosted on heroku here (it may be a bit slow to load since it is free-tier)
An image scraper version of the Flask app is located on the flaskv2 branch.
Summer project that trained a Pytorch classifier with a dataset of tens of thousands of photos. The model uses resnet50 as a base and via transfer learning utilizes a hand labeled dataset of about 1500 vacation photos to be better biased towards the inputs. I used a skeleton code for multitag pytorch training as well as a flask-heroku web app skeleton code and modified it to suit my needs. I am still tuning the model's accuracy so further commits updating 'model.pth' will occur on a regular basis.
Install them from requirements.txt:
pip install -r requirements.txt
Run the server:
python app.py
The mighty MIT license. Please check LICENSE for more details.