magic/zerocap

Our Implementation of the ZeroCap Baseline Model

---

Catalogue:

• 1. Environment Preparation
• 2. Image Captioning on MSCOCO
• 3. Image Captioning on Flickr30k
• 4. Cross Domain Image Captioning on MSCOCO
• 5. Cross Domain Image Captioning on Flickr30k
• 6. Citation
• 7. Acknowledgements

---

1. Environment Preparation:

To install the correct environment, please run the following command:

pip install -r requirements.txt

---

2. Image Captioning on MSCOCO:

To perform image captioning on MSCOCO, please run the following command:

chmod +x ./mscoco_zerocap.sh
./mscoco_zerocap.sh

---

3. Image Captioning on Flickr30k:

To perform image captioning on Flickr30k, please run the following command:

chmod +x ./flickr30k_zerocap.sh
./flickr30k_zerocap.sh

---

4. Cross Domain Image Captioning on MSCOCO:

To perform image captioning on MSCOCO with the language model from Flickr30k domain, please run the following command:

chmod +x ./flickr30k_to_mscoco_zerocap.sh
./flickr30k_to_mscoco_zerocap.sh

---

5. Cross Domain Image Captioning on Flickr30k:

To perform image captioning on Flickr30k with the language model from MSCOCO domain, please run the following command:

chmod +x ./mscoco_to_flickr30k_zerocap.sh
./mscoco_to_flickr30k_zerocap.sh

---

6. Citation:

If you find our code helpful, please cite the original paper as

@article{tewel2021zero,
  title={Zero-Shot Image-to-Text Generation for Visual-Semantic Arithmetic},
  author={Tewel, Yoad and Shalev, Yoav and Schwartz, Idan and Wolf, Lior},
  journal={arXiv preprint arXiv:2111.14447},
  year={2021}
}

---

7. Acknowledgements:

We thank the authors for releasing their code. Our reimplementation of the baseline is based on their original codebase [here].