📦 COinCO-resources

This repository contains all necessary preprocessed resources and pretrained models required to run the code for the COinCO. It supports the following downstream tasks:

1. In- and out-of-context classification
2. Objects-from-Context Prediction
3. Context-empowered fake localization

---

📁 Repository Contents

After cloning, you will find the following files:

• checkpoints.zip:
  Pretrained model checkpoints for downstream tasks such as context classification, object prediction, and fake localization.

• task_data_part_aa, task_data_part_ab, task_data_part_ac:
  These are split parts of task_data.zip, which contains:

  • Preprocessed data required to run all code modules
  • Baseline prediction results for the fake localization task (used for context enhancement and evaluation)

• README.md:
  Instructions for unpacking and using the resources.

---

🔧 How to Use

1. Clone the Dataset

git clone https://huggingface.co/datasets/ytz009/COinCO-resources
cd COinCO-resources

2. Reconstruct the Task Data

Concatenate the split archive files and unzip:

cat task_data_part_* > task_data.zip
unzip task_data.zip

You will now have a new folder task_data/ containing all required inputs and intermediate results.

3. Unzip Pretrained Checkpoints

unzip checkpoints.zip

---

📌 Notes

• The task_data folder is essential to run the code without needing to regenerate or recompute intermediate files.
• The pretrained models in checkpoints were trained using the official COinCO training set and are directly usable for evaluation or fine-tuning.

---

📚 Citation

If you use this dataset or resource package, please cite the accompanying paper:

Common Inpainted Objects In-N-Out of Context
Tianze Yang*, Tyson Jordan*, Ninghao Liu, Jin Sun
Submitted to NeurIPS 2025 Datasets and Benchmarks Track

---