FDP

This is the implementation of the paper "Focus, Distinguish, and Prompt: Unleashing CLIP for Efficient and Flexible Scene Text Retrieval".

Usage

First, install PyTorch 1.7.1 (or later) and torchvision, and install the CLIP as a Python package.

$ conda install --yes -c pytorch pytorch=1.7.1 torchvision cudatoolkit=11.0
$ pip install ftfy regex tqdm
$ pip install git+https://github.com/openai/CLIP.git

Then, install small additional dependencies.

$ pip install -r requirements.txt

Datasets

The MLT dataset can be download from here

The PSTR (Phrase-level Scene Text Retrieval) dataset can be download from here