yusasif/Nigerian-food-recognition-v2
yusasif/Nigerian-food-recognition-v2 is a 12 billion parameter Gemma 4 vision model developed by yusasif, fine-tuned to accurately identify over 165 Nigerian food classes from images. This model excels at precisely naming Nigerian dishes, including various swallows, soups, rice dishes, and snacks, making it highly specialized for food recognition within a specific cultural context. It leverages LoRA fine-tuning on both vision and language layers for efficient training and performance.
Loading preview...
Overview
yusasif/Nigerian-food-recognition-v2 is a specialized vision model built on the Gemma 4 12B architecture, fine-tuned to identify a wide array of Nigerian food dishes from images. Developed by yusasif, this model is designed to provide exact names for Nigerian foods, distinguishing it from general-purpose image recognition models.
Key Capabilities
- Highly Specialized Food Recognition: Identifies over 165 distinct Nigerian food classes, including various swallows (Eba, Iyan), soups (Egusi, Ogbono), rice dishes (Jollof rice, Ofada rice), snacks (Suya, Puff Puff), and regional specialties.
- Vision-Language Integration: Fine-tuned with LoRA on both vision and language layers using Unsloth for memory-efficient training, enabling robust image-to-text understanding.
- Comprehensive Dataset: Trained on approximately 600 image-text pairs, enriched with detailed Gemma-generated captions, ensuring a broad and accurate understanding of Nigerian cuisine.
- Efficient Deployment: The final checkpoint is merged to float16 and can be optionally quantized to 8-bit for serving, with vLLM support for an OpenAI-compatible API.
Good For
- Identifying Nigerian Dishes: Ideal for applications requiring precise identification and naming of Nigerian foods from visual input.
- Cultural Food Recognition: Useful for projects focused on cultural heritage, culinary education, or dietary tracking within the Nigerian context.
- Research and Development: Provides a strong baseline for further research into specialized food recognition models or for integrating into larger culinary AI systems.