task.md

Machine Learning Engineer Test Task: Text2Image Search

Objective

This test task aims to evaluate your grasp of fundamental machine learning and development concepts. Your task involves working with an image dataset to develop a system capable of searching for similar images based on textual queries.

Dataset

Please choose whatever image dataset you prefer, for example:

• crawl some e-commerce website
• find something on Kaggle
• use prepared Advertisement Image Dataset

#!/bin/bash

wget https://storage.googleapis.com/ads-dataset/subfolder-0.zip
wget https://storage.googleapis.com/ads-dataset/subfolder-1.zip

Task Requirements

Data Exploration

• Provide summarized report of the dataset contents

Text2Image Search Implementation

• Implement a service, capable of retrieving relevant images based on the text input provided.
  • For example: Text query "black cat" should return a picture of a black cat if it is present in the dataset

Tips:

• Use vector search with Qdrant
• Provide solution in a form of FastAPI service or streamlit application

Query Evaluation

• Provide a set of example queries that effectively showcase the capabilities of your implemented system. These queries should yield relevant images, demonstrating the accuracy and efficiency of your text2image search solution.
• Provide examples of queries that do not perform well, accompanied by explanations outlining the shortcomings of the system.
• Suggest a method of quantitive evaluation of retrieval accuracy. (e.g. how to label dataset and prepare queries?)

Submission Guidelines

• Upload your solution to your GitHub account, accompanied by clear instructions for running the text2image search system.
• Include a README.md describing the system's architecture and techniques employed, challenges encountered during the implementation, and potential avenues for improvement.

Please don't share links to the solution here