Global and Local Entailment Learning for Natural World Imagery

Srikumar Sastry*, Aayush Dhakal, Eric Xing, Subash Khanal, Nathan Jacobs (*Corresponding Author)

ICCV 2025

Radial Cross-Modal Embeddings (RCME) is a state-of-the-art hierarchical image-text ordering and retrieval method in the embedding space.

🔥 Textual Entailment and Ordering

📦 Models

Model	Architecture	HuggingFace
CLIP	ViT-B/16	MVRL/rcme-vit-base-patch16
CLIP	ViT-L/14	MVRL/rcme-vit-large-patch14
TreeofLife	ViT-B/16	MVRL/rcme-tol-vit-base-patch16

⚙️ Setup

Setting up

1. Clone this repository:

git clone https://github.com/mvrl/RCME.git

2. Install dependencies:

cd RCME && pip install -r requirements.txt

🗂️ Data

1. Use BioCLIP's scripts to download TreeofLife-10M dataset:

rcme/data/bioclip/scripts/setup_download_tol-10m_components.bash && \
rcme/data/bioclip/scripts/submit_download_tol-10m_components.bash

Hint: Setup paths and other variables in setup_download_tol-10m_components.bash script.

2. Use our script to convert TreeofLife-10M dataset into iNaturalist-2021 style naming:

python rcme/data/bioclip/write_imgs.py

Hint: Setup paths and other variables in our script.
Hint: Currently only supports num_workers=1

🔥 Training

1. Setup all hyperparameters in rcme/config.py file.
2. Run training by specifying the model:

python rcme/train.py --model="rcme"

Hint: Currently supports rcme, radial, atmg and meru.

✅ Evaluation

Scripts and documentation coming soon...

📑 Citation

@inproceedings{sastry2025global,
    title={Global and Local Entailment Learning for Natural World Imagery},
    author={Sastry, Srikumar and Dhakal, Aayush and Xing, Eric and Khanal, Subash and Jacobs, Nathan},
    booktitle={International Conference on Computer Vision},
    year={2025},
    organization={IEEE/CVF}
}

🔍 Additional Links

Check out our lab website for other interesting works on geospatial understanding and mapping:

• Multi-Modal Vision Research Lab (MVRL) - Link
• Related Works from MVRL - Link