藉由物件測定與深層卷積網路辨識流行服飾

隨著社群網路上照片和視頻分享的流行，分析使用者照片上的人類屬性，包括人臉屬性、身體形狀、以及服裝的屬性，成為時下許多有用的應用程式的重要問題。此外，隨著網絡購物平台的興盛，分析使用者照片的服裝屬性成為時下相關應用程式的一個重要關鍵。在本文中，我們提出了一個方法來辨識服裝屬性中最重要的服裝類別，我們從各種不同平台匯集了多樣化的資料集，並利用已顯示能運用在各種領域，有能力解決各種問題的深層卷積網路。過去的研究主要利用以部分為基礎的方法（例如，偵測身體部位，軀幹四肢估計）作為定位找到服裝相關的區域來進行服裝的辨識。然而，這種方法本身限制了這個問題，以致於只能做在以人為中心的照片。在這份研究中，我們的方法除去照片中有人的基本假設，並學習自動化地識別真實照片中的多個服裝的類別。

關鍵字

深度學習；完全卷積網路；服裝辨識；多重標籤預測；自動化標記

並列摘要

Recently, with the emerging of online shopping platforms, analyzing clothing attributes on user-generated photos becomes an important key for the related applications nowadays. In this paper, we introduce a system framework for recognizing fashion apparel (i.e., clothing categories), which is the crucial part of clothing attributes. We integrate the datasets with large variety collected from different platforms and utilize the powerfulness of Deep Convolutional Neural Network which has been shown the capability of adapting to various domains and tasks. Previous works mainly use the part-based methods (e.g., body part detector, pose estimator) as alignment to locate the clothing-related regions for recognizing clothing. However, this constraints the problem on human centric photos. In this work, our proposed method removes the constraint of human body existence which is generally used for alignment and learns with image-level annotated data to recognize multiple clothing garments in real-world data.

並列關鍵字

Deep learning ； Fully convolutional network ； clothing recognition ； multi-label prediction ； automatical annotation

參考文獻

compared to state-of-the-art superpixel methods. Pattern Analysis and Machine

[2] L. Bourdev, S. Maji, and J. Malik. Describing people: A poselet-based approach

[3] L. Bourdev and J. Malik. Poselets: Body part detectors trained using 3d human

on, pages 1365–1372. IEEE, 2009.

[4] H. Chen, A. Gallagher, and B. Girod. Describing clothing by semantic attributes. In

國際替代計量

藉由物件測定與深層卷積網路辨識流行服飾

全文下載

主題瀏覽