Image Recognition

Scraping, indexing and classification with a Telegram Bot

This project scrapes images from Flickr, indexes them into Elasticsearch, and trains image classifiers (HOG + SVM). It also includes a Telegram bot for inference.

Requirements

• Linux OS or virtual machine
• Python 3.9 (conda environment recommended)
• Reachable Elasticsearch (default https://localhost:9200)
• Chrome/Chromedriver installed

Environment setup

1. Create the conda env using img_class.yml: conda env create -f img_class.yml
2. Copy .env.example to .env and fill in the values:

Main .env variables:

• TELEGRAM_TOKEN: your Telegram bot token
• ELASTIC_HOST: Elasticsearch URL (e.g. https://localhost:9200)
• ELASTIC_PASSWORD: password for elastic user
• ELASTIC_CA_CERT: path to CA certificate if applicable (optional)
• DRIVER_PATH: path to chromedriver
• DOWNLOADS_PATH: root folder for images (e.g. Imagenes)
• FLICKR_USERNAME, FLICKR_PASSWORD: Flickr credentials (optional, if logging in)
• MODELS_PATH: models folder (e.g. Modelos)

3. Add a driver_path folder with the chromedriver executable.

Project structure

• main.py: orchestrates scraping, split, indexing and training
• bot.py: Telegram bot for classification
• scraping.py: Selenium helpers (search and download)
• indexation.py: Elasticsearch connection and indexing
• model.py: model training and evaluation
• config.py: environment variable loader
• Modelos/: saved models
• Imagenes/: dataset (train/test/backup)

Run

1. Execute the full pipeline:

python main.py

2. Start the bot (after indices and models exist):

python bot.py

Security notes

• Do not commit .env or any credentials.
• Ensure ELASTIC_CA_CERT points to an accessible certificate if needed.