Clothing Type and Color Classification Using TinyML

This paper introduces an edge-AI system designed to assist visually impaired individuals in achieving greater independence in clothing selection by specifically detecting both the type and color of the garments. We use embedded machine learning within a user centric approach to bring together TinyML and intelligent devices for a real-time cloth and color detection system. We finetune MobileNetV2 on a clothing dataset curated from fashion cloth images. The results show that the model achieved an accuracy of 89% and an F1-score of 88%. We also considered edge deployment to make the solution available real-time in tiny devices for the visually impaired people. We use post-training quantization to reduce the model's size and evaluate deployment options for Arduino Nano 33 BLE Sense (Cortex-M4F 64MHz). The quantized model achieves a minimal latency of 177 milliseconds, requiring only 60KB of RAM and 109KB of storage memory, making it an optimal choice for deployment on our tiny device.