Fine-grained Image Retrieval

This is the Official Pytorch-Lightning implementation of the paper: "Cascading Hierarchical Networks with Multi-task Balanced Loss for Fine-grained hashing"(https://arxiv.org/abs/2303.11274).

Requirements

• Python3
• PyTorch
• PyTorch-Lightning
• Numpy
• pandas

---

Dataset

We use the following 5 datasets: CUB200-2011, Aircraft, VegFru, NABirds and Food101.

---

RUN

• The running commands for several datasets are shown below.

python main.py --dataset cub --gpu 0, --batch_size=128 --code_length=32 --num_workers=4  --lr 0.008
python main.py --dataset aircraft --gpu 0, --batch_size=32 --code_length=32 --num_workers=4 --lr 0.035
python main.py --dataset vegfru --gpu 0, --batch_size=128 --code_length=32 --num_workers=4  --lr 0.005
python main.py --dataset nabirds --gpu 0, --batch_size=128 --code_length=32 --num_workers=4  --lr 0.008
python main.py --dataset food101 --gpu 0, --batch_size=128 --code_length=32 --num_workers=4  --lr 0.008

---

Citation

If you find our work inspiring or use our codebase in your research, please cite our work.

@misc{zeng2023cascading,
      title={Cascading Hierarchical Networks with Multi-task Balanced Loss for Fine-grained hashing}, 
      author={Xianxian Zeng and Yanjun Zheng},
      year={2023},
      eprint={2303.11274},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}