OneScience-Group/RemoteCLIP

RemoteCLIP

Model Introduction

RemoteCLIP is a vision-language foundation model for remote sensing imagery and text. It aligns remote sensing visual semantics with natural-language descriptions through CLIP dual encoders and bidirectional contrastive learning, supporting cross-modal retrieval and transfer to downstream remote sensing tasks.

Paper: RemoteCLIP: A Vision Language Foundation Model for Remote Sensing
https://arxiv.org/abs/2306.11029

Model Description

RemoteCLIP was proposed by a research team from the National University of Defense Technology and other institutions. The model performs continual pretraining with RSITMD, RSICD, UCM-Captions, and remote sensing image-text data constructed by converting task datasets. It is suitable for remote sensing image-text retrieval, zero-shot classification, and vision-language representation learning.

Use Cases

Scenario	Description
Remote sensing image-text retrieval	Compute cross-modal similarities between image and text features.
Multi-positive contrastive learning	Use pair_ids to express positive relationships such as one image paired with multiple texts.
Zero-shot scene classification	Match natural-language class prompts with image features to recognize remote sensing scenes without training an additional classification head.
Few-shot visual recognition	Transfer vision-language representations and perform few-shot fine-tuning or linear probing for remote sensing classification with limited labels.
Local engineering validation	Use a small amount of synthetic data to check the training, inference, and evaluation workflows.
Multi-GPU training	Launch distributed training with torchrun.

Usage Guide

1. OneCode

Experience intelligent one-click AI4S programming through the OneCode online environment:

Click to Experience Intelligent One-Click AI4S Programming

2. Download and Installation

hf download OneScience-Group/RemoteCLIP --local-dir ./RemoteCLIP
cd RemoteCLIP

Environment Dependencies

Hardware Requirements

• A GPU or DCU is recommended.
• CPU can be used for small-configuration connectivity validation; full training and inference will be slow.
• DCU users must install DTK in advance. DTK 25.04.2 or above, or the OneScience recommended version matching the current cluster, is recommended.

DCU Environment

# Please activate DTK and CONDA first
conda create -n onescience311 python=3.11 -y
conda activate onescience311
# uv installation is supported
pip install onescience[earth-dcu] -i http://mirrors.onescience.ai:3141/pypi/simple/ --trusted-host mirrors.onescience.ai

GPU Environment

# Please activate CONDA first
conda create -n onescience311 python=3.11 -y libstdcxx-ng=12 libgcc-ng=12 gcc_linux-64=12 gxx_linux-64=12
conda activate onescience311
# uv installation is supported
pip install onescience[earth-gpu] -i http://mirrors.onescience.ai:3141/pypi/simple/ --trusted-host mirrors.onescience.ai

Training Data Introduction

By default, 8 training and 4 test synthetic image-text samples are used to validate the engineering workflow. They are saved as data/train.npz and data/test.npz, respectively. Tokens follow the OpenAI CLIP BPE vocabulary of 49,408 entries and the SOT, EOT, and padding sequence constraints.

The synthetic data preserves the official model input specifications of 3-channel 224x224 images and CLIP text sequences of length 77.

Real data must be preprocessed and converted to the following NPZ training protocol. This protocol is consistent with the model input specification but is not the download format of the original datasets.

images: float32 [N,3,224,224]
tokens: int64 [N,77]
pair_ids: int64 [N]

fake_data.py automatically writes the protocol and data_source protocol metadata. These fields must be retained when using real data.

python scripts/fake_data.py

Training

python scripts/train.py

For multi-GPU training, use:

torchrun --nproc_per_node=8 scripts/train.py

Training optimizes the image and text dual encoders with a multi-positive bidirectional contrastive objective and saves a checkpoint and aggregate training metrics. The default configuration is intended for quick workflow validation. Formal experiments should use the image-text data scale, model configuration, and training duration corresponding to the paper.

result/checkpoints/remoteclip.pt
result/training/metrics.json

Training Weights

This repository will provide RemoteCLIP training weights in the weight/ folder. The weight files will be uploaded soon and are expected to be available in the near future.

Inference

python scripts/inference.py

Inference loads the training checkpoint, computes test-set image and text features and their similarities, and saves the results to:

result/output/retrieval.npz

Evaluation and Visualization

python scripts/result.py

Using pair_ids, evaluation reports bidirectional retrieval R@1, R@5, R@10, and mean recall, and generates a similarity heatmap. Results on synthetic data are only for engineering workflow validation and do not represent full-paper performance.

result/evaluation/metrics.json
result/evaluation/similarity_matrix.png

Official OneScience Resources

Platform	OneScience Main Repository	Skills Repository
Gitee	https://gitee.com/onescience-ai/onescience	https://gitee.com/onescience-ai/oneskills
GitHub	https://github.com/onescience-ai/OneScience	https://github.com/onescience-ai/oneskills

Citation and License

This repository is a reproduction of the original RemoteCLIP paper.

Use of the code and data in this repository remains subject to the licenses and terms of use of their respective projects.