CSTBIR

Code from paper: Composite Sketch+Text Queries for Retrieving Objects with Elusive Names and Complex Interactions

project page | paper

Requirements

• Use python >= 3.8.16. Conda recommended : https://docs.anaconda.com/anaconda/install/linux/

• Use pytorch 1.13.1 CUDA 11.6

• Other requirements from 'requirements.txt' and 'environment.yaml'

To setup environment

# create new env cstbir
$ conda create -n cstbir python=3.8.16

# activate cstbir
$ conda activate cstbir

# install other dependencies
$ conda env update --file environment.yml --prune
$ pip install -r requirements.txt

Preparing dataset

• Download VG images from https://homes.cs.washington.edu/~ranjay/visualgenome/index.html

• Download QuickDraw Sketches from https://github.com/googlecreativelab/quickdraw-dataset

• Download CSTBIR dataset from Google Drive Link

Store the downloaded dataset in the ./data/ directory.

Running the code

Training parameters

To check and update training, model and dataset parameters see configs

To train the model:

$ CUDA_VISIBLE_DEVICES=XX python run.py

Cite

If you find this code/paper useful for your research, please consider citing.

@InProceedings{cstbir2024aaai,
        author    = {Gatti, Prajwal and Parikh, Kshitij Gopal and Paul, Dhriti Prasanna and Gupta, Manish and Mishra, Anand},
        title     = {Composite Sketch+Text Queries for Retrieving Objects with Elusive Names and Complex Interactions},
        booktitle = {AAAI},
        year      = {2024},
}   

Acknowledgements

This work uses https://github.com/openai/CLIP/ for the implementation of the CLIP model.