imgurer

A functioning image repository API and frontend with upload and similar(duplicate) image search capabilities.

installation

NOTE:

• using virtualenvwrapper is recommended
• currently environment variables are required
  • --> defaults are supplied in postactivate.txt

create a virtual environment in current directory
activate virtual environment 



NOTE: using virtualenvwrapper is recommended
  create environment variables as per postactivate.txt (unset in predeactivate.txt)
  https://virtualenvwrapper.readthedocs.io/en/latest/scripts.html?highlight=postactivate#postactivate



copy repository
install requirements
(optionally) "chmod +x run" to >more easily start server
run server and navigate to >localhost (127.0.0.1:8000/)
127.0.0.1:8000/docs for directly interfacing with the API

Python only (no venvwrapper) installation using bash

python3 -m venv './imgurer'
source bin/activate
git clone https://github.com/zzpz/imgurer.git
cd imgurer

set -o allexport
source postactivate.txt
set +o allexport

pip install -r requirements
uvicorn imgurer.main:app --reload

requirements

imgurer will currently write files to local disk, if this is restricted expect errors virtualenvwrapper allows for pre and post activation hooks (setting env variables such as $DB_URL)

Usage

• The application is accessible through both API calls and a frontend on local host.
• User creation is only accessible through the /docs API backend

Future areas of development

• refactor to external db (:elephant: sql)
• refactor of image storage from local to CDN/NAS
• external image store(s)
• user login and cookied credentials
• frontend framework (react/vue/etc)
• parsing image exif data on upload
• migration to postgres
• extending models to link user and images

notes

• images uploaded are stored on disk as both a thumbnail and the original

  • be mindful of filling a harddrive while uploading

• on server reload the ability to search files already uploaded is lost

  • accessing </images/rebuildBKT> will rebuild the search tree
  • the tree can be rebuilt from the search page

• search for similarity is done via a 128bit dhash comparison

  • this is more accurately a search for 'duplicate' images
  • a more sophisticated comparison should then be used for less false positives

• file uploads are not validated

  • size is not checked in headers
  • files are not chunked for upload

• images are bucketed to multiple folders with a hashed name to allow for enormous image sets

• frontend javascript and html is a mess and little to none is re-used

• this is in no way safe for a production environment