Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Privacy Preserving Ordinal-Meta Learning with VLMs for Fine-Grained Fruit Quality Prediction

Riddhi Jain, Manasi Patwardhan, Aayush Mishra, Parijat Deshpande, Beena Rai

TL;DR

The paper tackles privacy-preserving, fine-grained fruit freshness prediction from visual data under scarce labeling. It introduces Model-Agnostic Ordinal Meta-Learning (MAOML), which blends meta-learning (MAML) with CORN-style ordinal regression to train small open-source Vision-Language Models on ordinal fruit-quality labels. MAOML achieves state-of-the-art-like performance in zero-shot and few-shot settings, with reported accuracies of 90.28% (zero-shot) and 92.71% (few-shot), often outperforming a large proprietary model Gemini. This approach enables on-site, privacy-conscious deployment in food retail, reducing waste while maintaining high accuracy with limited labeled data.

Abstract

To effectively manage the wastage of perishable fruits, it is crucial to accurately predict their freshness or shelf life using non-invasive methods that rely on visual data. In this regard, deep learning techniques can offer a viable solution. However, obtaining fine-grained fruit freshness labels from experts is costly, leading to a scarcity of data. Closed proprietary Vision Language Models (VLMs), such as Gemini, have demonstrated strong performance in fruit freshness detection task in both zero-shot and few-shot settings. Nonetheless, food retail organizations are unable to utilize these proprietary models due to concerns related to data privacy, while existing open-source VLMs yield sub-optimal performance for the task. Fine-tuning these open-source models with limited data fails to achieve the performance levels of proprietary models. In this work, we introduce a Model-Agnostic Ordinal Meta-Learning (MAOML) algorithm, designed to train smaller VLMs. This approach utilizes meta-learning to address data sparsity and leverages label ordinality, thereby achieving state-of-the-art performance in the fruit freshness classification task under both zero-shot and few-shot settings. Our method achieves an industry-standard accuracy of 92.71%, averaged across all fruits. Keywords: Fruit Quality Prediction, Vision Language Models, Meta Learning, Ordinal Regression

Privacy Preserving Ordinal-Meta Learning with VLMs for Fine-Grained Fruit Quality Prediction

TL;DR

The paper tackles privacy-preserving, fine-grained fruit freshness prediction from visual data under scarce labeling. It introduces Model-Agnostic Ordinal Meta-Learning (MAOML), which blends meta-learning (MAML) with CORN-style ordinal regression to train small open-source Vision-Language Models on ordinal fruit-quality labels. MAOML achieves state-of-the-art-like performance in zero-shot and few-shot settings, with reported accuracies of 90.28% (zero-shot) and 92.71% (few-shot), often outperforming a large proprietary model Gemini. This approach enables on-site, privacy-conscious deployment in food retail, reducing waste while maintaining high accuracy with limited labeled data.

Abstract

To effectively manage the wastage of perishable fruits, it is crucial to accurately predict their freshness or shelf life using non-invasive methods that rely on visual data. In this regard, deep learning techniques can offer a viable solution. However, obtaining fine-grained fruit freshness labels from experts is costly, leading to a scarcity of data. Closed proprietary Vision Language Models (VLMs), such as Gemini, have demonstrated strong performance in fruit freshness detection task in both zero-shot and few-shot settings. Nonetheless, food retail organizations are unable to utilize these proprietary models due to concerns related to data privacy, while existing open-source VLMs yield sub-optimal performance for the task. Fine-tuning these open-source models with limited data fails to achieve the performance levels of proprietary models. In this work, we introduce a Model-Agnostic Ordinal Meta-Learning (MAOML) algorithm, designed to train smaller VLMs. This approach utilizes meta-learning to address data sparsity and leverages label ordinality, thereby achieving state-of-the-art performance in the fruit freshness classification task under both zero-shot and few-shot settings. Our method achieves an industry-standard accuracy of 92.71%, averaged across all fruits. Keywords: Fruit Quality Prediction, Vision Language Models, Meta Learning, Ordinal Regression

Paper Structure

This paper contains 13 sections, 3 equations, 1 figure, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Fruit freshness detection
  4. Meta-Learning Vision Language Models
  5. Dataset
  6. Methods
  7. In-context Learning
  8. Fine-tuning
  9. Meta-Learning
  10. Model Agnostic Ordinal-Meta Learning (MAOML)
  11. Experiments
  12. Results and Discussion
  13. Conclusion

Figures (1)

  • Figure 1: Illustration of the need of an on-site model for fine-grained fruit freshness classification without compromising on the performance achieved by Large proprietary models. Our baselines and expected outcomes by our approach of Model Agnostic Ordinal Meta-Learning (MAOML).