Abstract:Clothing detection is a hot research focus as its application of identifying the specific category of clothing, such as long-sleeved and short-sleeved. Image-based clothing detection requires the model to detect accurate position. At present, the approaches of clothing detection are mainly divided into two categories: one is top-down, which is anchor-based and needs to calculate the intersection over union between the anchor box and the bounding box, but it is limited by the setting of the anchor box and does not perform well when the clothing scale is variable; the other is bottom-up, which uses the feature extraction network to get the keypoints and calculates the position and size of the clothing, but the prediction of the keypoints often has a slight error for it lacks the internal information of the clothing. To address the above issues, we propose a multi-keypoints matching network for clothing detection (MKMnet) based on the bottom-up method. It detects three keypoints (top-left corner point, bottom-right corner point, and center point) to ensure high detecting accuracy. Firstly, we perform corner keypoint matching by calculating the distance between the embedding vectors of different corner points to get the initial bounding box, then we get the final bounding box by matching the center point. The way to get the bounding box by corner point matching makes the model have the ability to detect clothing of any scale and shape, and adding the center point for further verification eliminates a large number of false-positive bounding boxes. The MKMnet proposed in this paper can obtain the bounding boxes accurately through the linear combination of the center point, and improving the accuracy of clothing recognition. The experimental results show that the MKMnet has higher accuracy than existing methods.

Multi-depth Dilated Network for Fashion Landmark Detection with Batch-Level Online Hard Keypoint Mining.

Clothing Landmark Detection Using Deep Networks With Prior of Key Point Associations

Hierarchical Multi-Scale Network for Cross-Scale Visual Defect Detection

Multiple-Clothing Detection and Fashion Landmark Estimation Using a Single-Stage Detector

Real-Time Facial Landmark Detection by Attention-driven Lightweight Network

Unconstrained Fashion Landmark Detection via Hierarchical Recurrent Transformer Networks

Multi-keypoints matching network for clothing detection

Fashion Landmark Detection in the Wild

Improving Fashion Landmark Detection by Dual Attention Feature Enhancement.

Deep Fashion Analysis with Feature Map Upsampling and Landmark-Driven Attention

Adaptive Graph Reasoning Network for Fashion Landmark Detection

Spatial-Aware Non-Local Attention for Fashion Landmark Detection

Deep Deformation Network for Object Landmark Localization

Two-Stream Multi-Task Network for Fashion Recognition

A Global-Local Emebdding Module for Fashion Landmark Detection

Robust Facial Landmark Detection by Multi-order Multi-constraint Deep Networks

PCLoss: Fashion Landmark Estimation with Position Constraint Loss.

DeepMark++: Real-time Clothing Detection at the Edge

Clothes Keypoints Localization and Attribute Recognition Via Prior Knowledge.

A Deep-Learning-Based Fashion Attributes Detection Model