Scholarlink

<p>Color classification is a critical task across various industries. Deep learning color classification methods are accurate but have high computational complexity, which limits their use in real-world applications. We presented a pipeline employing a lightweight classification architecture to tackle this issue effectively. The tailored architecture was developed based on the EfficientNet-B0 model, integrating innovative residual connections to improve the extraction of color features. This approach resulted in an exceptionally lightweight model with only 1.90 million parameters and a Giga Floating Point Operations per Second (GFLOPs) count of 0.03 while still delivering outstanding at 94.96% and 78.06% for public vehicle and avocado datasets. We also introduced a new benchmark dataset specifically for Vietnamese motorbike color classification and achieved an accuracy of 91.36%. Our code can be accessed at <a href="https://github.com/taipt03/MoCo/tree/main">https://github.com/taipt03/MoCo/tree/main</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CoNet: a lightweight color classification architecture using residual connection and MBConv

  • 期刊论文

  • 发表日期：

    2025年3月6日

查看全文
  • Tien Dung Bui,
  • Tuan Tai Pham,
  • Tuan Linh Dang

摘要

Color classification is a critical task across various industries. Deep learning color classification methods are accurate but have high computational complexity, which limits their use in real-world applications. We presented a pipeline employing a lightweight classification architecture to tackle this issue effectively. The tailored architecture was developed based on the EfficientNet-B0 model, integrating innovative residual connections to improve the extraction of color features. This approach resulted in an exceptionally lightweight model with only 1.90 million parameters and a Giga Floating Point Operations per Second (GFLOPs) count of 0.03 while still delivering outstanding at 94.96% and 78.06% for public vehicle and avocado datasets. We also introduced a new benchmark dataset specifically for Vietnamese motorbike color classification and achieved an accuracy of 91.36%. Our code can be accessed at https://github.com/taipt03/MoCo/tree/main.