Multimodal-Outpost-Notebooks

Multimodal-Outpost is a collection of Colab notebooks designed for image inference and multimodal vision-language model (VLM) experimentation. It provides tools for OCR, image captioning, video understanding and generating DOCX or PDF documents containing both images and extracted text.

Notebooks List 📘

This repository contains a curated collection of notebooks for implementing state-of-the-art multimodal Vision-Language Models (VLMs).

Notebook Name	Link ↗
Aya-Vision-8B-VideoUnderstanding	Link
Behemoth-3B-070225-post.1	Link
Camel-Doc-OCR-080125	Link
Florence-2-Models-Image-Caption	Link
Gemma3-VL-VideoUnderstanding	Link
Imgscope-OCR-2B-0527-VideoUnderstanding	Link
Inkscope-Captions-2B-0526-VideoUnderstanding	Link
LFM2-VL-1.6B-LiquidAI	Link
LFM2-VL-450M-LiquidAI	Link
Lumian-VLR-7B-Thinking-Demo-Notebook	Link
Lumian2-VLR-7B-Thinking-Demo-Notebook	Link
Megalodon-OCR-Sync-0713-ColabNotebook	Link
MiMo-VL-7B-RL-VideoUnderstanding	Link
MiMo-VL-7B-SFT-VideoUnderstanding	Link
MonkeyOCR-0709	Link
OCRFlux3B	Link
Qwen2-VL-MessyOCR-VideoUnderstanding	Link
Qwen2-VL-OCR-2B-Instruct	Link
Qwen2-VL-VideoUnderstanding	Link
Qwen2.5-VL-3B-Abliterated-Caption-it(caption)	Link
Qwen2.5-VL-3B-Instruct	Link
Qwen2.5-VL-7B-Abliterated-Caption-it	Link
Qwen2.5-VL-VideoUnderstanding	Link
RolmOCR-Qwen2.5-VL-VideoUnderstanding	Link
SmolDocling-256M-preview	Link
monkey-OCR	Link
moondream2-2025-06-21	Link
nanonets-OCR	Link
olmOCR-Qwen2-VL-VideoUnderstanding	Link
typhoon-OCR	Link
typhoon-ocr-7b-Qwen2.5VL-VideoUnderstanding	Link

Features

• Extracts text from images using various OCR models
• Supports image captioning and multimodal inference
• Embeds images and extracted text into DOCX or PDF formats
• Designed for quick deployment via Google Colab

Inference Image (Demo)

Dependencies

• Python
• PyTorch
• Hugging Face Transformers
• ReportLab
• Gradio (for UI)
• (Qwen2.5-VL based) / Others

All dependencies are automatically installed in the Colab environment.

Author

Created and maintained by PRITHIVSAKTHIUR