Person retrieval, using attire color and type has gained significant attention recently due to its potential security, surveillance, and crowd management applications. Unlike biometric data, attire color and type information is not personally identifiable and can be easily anonymized. This privacy-preserving aspect makes person retrieval using cloth color more desirable in certain contexts. Our work focused on identifying children by analyzing the various types of dresses and their colors in our custom datasets. Here, we used 12 colors and 11 types of attire on the upper and lower body. Upper body clothing options include T-shirts, full sleeves, half sleeves, kurtas, and coats. Clothing for the lower body includes jeans, pants, skirts, leggings, pyjamas and uniforms. Despite progress, precisely retrieving children remains difficult, especially owing to occlusion, illumination, angle, and scale. This paper introduces two significant contribution Child Adult Detection Network from the baselines of YOLOv8 that extracts essential feature maps using both local and global information. Also, we propose a multi-attribute recognition model called the Child Attire Retrieval Net (CARNet) that integrates various interdependencies of inter-classes to recognize all the attributes of child’s attire in a single image. Comparative study on state-of-the-art attire detection and retrieval frameworks is carried out. CADNet and CARNet achieves a mAP of 95.3% and 93.2% on the proposed dataset. In future, various other soft-biometric traits will be included for child detection and retrieval.