[paper]DeepFashion:Powering Robust Clothes Recognition and Retrieval with Rich Annotations
[Download Dataset]
DeepFashion:Powering Robust Clothes Recognition and Retrieval with Rich Annotaions
Motivation
Task: clothes recognition and retrieval(衣服识别和检索)
- Landmarks improve fine-grained recognition(改善细粒度的识别)
- Massive attributes(大量属性) better partition feature space(更好的划分要素空间)
- Photo pairs bridge the cross-domain gap(弥合跨域间距)
DeepFashion Dataset
-
Data Source(数据源)
Search engines(搜索引擎), online stores(网上商店), user posts(用户帖子). -
Quality Control(质量控制)
Duplicate removal(消除重复), fast screening(快速筛选), double checking(双重检查) -
Annotation Assessment(注释评估)
-
Sample Images(示例图像)
-
Category and Attribute Prediction Benchmark
-
In-shop Clothes Retrieval Benchmark
-
Consumer-to-shop Clothes Retrieval Benchmark
-
Fashion Landmark Detection Benchmark
-
-
Attributes Statistics(属性统计)
-
Landmarks and Pairs
FashionNet
-
Network Architecture
FashionNet jointly predicts landmarks and attributes to unify global and local feature learning.
-
Landmark Pooling Layer
Landmark pooling layer pools and gates features from estimated landmark locations.
-
Multi-task Learning
Cross-entropy loss for attributes, Euclidean loss for landmarks, triplet loss for pairs.
Benchmarks
-
Category & Attribute Prediction
-
In-shop Clothes Retrieval
-
Consumer-to-shop Clothes Retrieval