A feature addressable distributed hash table using vectors to search for similarity across a peer-to-peer network. Returning the most similar file. For a more detail read, check out the blog post
Note: running this on different machines has resulted in different output, likely a result of different Tensorflow versions
Tensorflow
Numpy
Pillow
Project currently in prototype stage, and what exists is the most basic implementation in a controlled environment.
python lookup.py [filepath]
The datalookup has DATA, which then has many directories as DATA/something/
This is an example that works, other files can still be off target.
python lookup.py 3311335910_36bf189ef6.jpg
There is no network interface right now, and DHT implementations that exist have hashing baked in. So a DHT needs to be implemented from scratch with the new primitives hardcoded in the codebase.
I have used a dht submodule from another repo