This repository contains a data processing project that involves data acquisition from Flickr, image conversion, caption generation, combining data into prompts, description generation, and seamless integration with Flickr for updates. It's a comprehensive solution for managing and enhancing visual content.
-
Data Acquisition via Flickr: Easily acquire images and data from Flickr using API. Download to Databricks volumes, can be modified to other paths.
-
Image Conversion: Convert images to PIL.
-
Caption Generation: Generate captions for your images using Salesforce/blip-image-captioning-large.
-
Combining Data into Prompts: Seamlessly combine acquired metadata into prompts for LLM model, mpt-30b-instruct/v1.
-
Description Generation: Generate descriptive content to enhance your visual assets and provide more context.
-
Integration with Flickr for Updates: Keep your content up-to-date on Flickr image description with a smooth API integration process.
- Start by configuring your Flickr API credentials within the project. I suggest using postman following : https://medium.com/apis-with-valentine/oauth-1-0-authorization-flow-using-flickr-api-and-postman-3-legged-oauth-c2a9b46bd8b9
- Set up a MosaicML api key, could be changed to a OpenAI model with code modification. Also possible to change to a hugging face model as well.
- Databricks set up with a cluster with a runtime: 13.3 LTS ML
Generate AI generated description to you public image on Flickr
- flickrapi
- PIL
- HuggingFace
- MosaicML
Feel free to add suggestion for improvements !