oceanir 0.1.0

Oculus Vision-Language Model - Inference SDK for multimodal AI research

Oceanir

Oculus Vision-Language Model SDK for multimodal AI research.

Installation

pip install oceanir

Quick Start

from oceanir import Oculus

# Load the model
model = Oculus.from_pretrained("OceanirAI/Oculus-0.1-Instruct")

# Visual Question Answering
answer = model.ask("photo.jpg", "What is the person doing?")
print(answer)  # "The person is riding a bicycle."

# Image Captioning
caption = model.caption("photo.jpg")
print(caption)  # "A dog playing in the park with a frisbee."

# Object Detection
results = model.detect("photo.jpg")
for box, label, conf in zip(results['boxes'], results['labels'], results['confidences']):
    print(f"{label}: {conf:.2f}")

# Counting Objects
count = model.count("crowd.jpg", "people")
print(f"Found {count} people")

Models

Model	Description
OceanirAI/Oculus-0.1-Instruct	Instruction-tuned for general VQA and captioning
OceanirAI/Oculus-0.1-Reasoning	Enhanced with chain-of-thought reasoning

Reasoning Mode

Enable thinking traces for complex questions:

# With reasoning
answer = model.ask(
    "complex_scene.jpg",
    "How many red cars are parked on the left side?",
    think=True
)

Features

• Visual Question Answering (VQA) - Answer questions about images
• Image Captioning - Generate natural language descriptions
• Object Detection - Detect and localize objects with bounding boxes
• Object Counting - Count specific objects in images
• Semantic Segmentation - Pixel-level scene understanding
• Chain-of-Thought Reasoning - Step-by-step reasoning for complex tasks

Architecture

Oculus combines:

• DINOv2 - Self-supervised vision transformer for semantic understanding
• SigLIP - Vision-language alignment for text understanding
• Trained Projector - Maps vision features to language space
• BLIP - Language model for text generation

License

This software is released under the Oceanir Research License.

Permitted Uses:

• Academic research
• Educational purposes
• Publishing papers with results
• Personal experimentation

Prohibited Uses:

• Commercial applications
• Training commercial models
• Integration into commercial products

For commercial licensing, contact: [e-mail]

Citation

If you use Oceanir in your research, please cite:

@software{oculus2026,
  title={Oculus Vision-Language Model},
  author={OceanirAI},
  year={2026},
  url={https://github.com/OceanirAI/oceanir}
}

Links

• HuggingFace Models
• Documentation
• GitHub