ECDM: an Efficient Clothing Detection Model Based on Optimized YOLOv8

The current detection system is faced with great challenges when dealing with various changes of clothing style, color and texture, especially when detecting fine details and small-size clothing. We propose a robust garment detection framework (ECDM) based on optimized yolov8 to solve this key challenge. This framework introduces a new c2flu dynamic module, and improves the original C2F by introducing conditional convolution. This enhancement increases the ability of our model to extract features more effectively. In order to further solve the problem of detecting clothing details and small-scale objects, ECDM introduces MSSFPN network with coordinated attention mechanism. This enhancement improves the ability of the model to focus on clothing features, enhances the multi-level feature fusion, and enables the model to better capture the outline and details of clothing. We used the deepfashion 2 dataset to evaluate ECDM and conducted a large number of experiments. The results show that ECDM has significant advantages in clothing detection, with high detection accuracy and efficiency. Compared with the advanced prediction model, this model has the best prediction performance mAP@50 and mAP@50: 95 increased by 8 % and 5 %, while model parameters, GFLOPs and size decreased by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$8 \%, 5 \%$</tex> and 23 %, respectively. This model can be obtained on GitHub https:stussy-449/ECDM (github.com)