--- title: "Baidu Brain and the Present and Future of the FMCG Industry" description: "On August 21, at the 2019 China FMCG Conference hosted by New Distribution, Xie Yongkang, Senior Technical Manager of Baidu AI Technology Ecosystem Department, introduced the overall architecture of Baidu Brain, Baidu's AI technology culmination, and shared the AI capabilities it currently provides for the FMCG industry, as well as its future plans." author: "谢永康" publisher: "New Distribution" email: "zhaobo258@gmail.com" telephone: "+8615854817671" published: "2019-09-09" categories: "Industry Trends" language: "en" canonical: "https://xinjignxiao.com/en/articles/baidu-brain-and-the-present-and-future-of-the-fmcg-industry-7d60d620/" markdown: "https://xinjignxiao.com/en/articles/baidu-brain-and-the-present-and-future-of-the-fmcg-industry-7d60d620.md" original_source: "https://mp.weixin.qq.com/s/goi0vuacMN18nih7qRZRnw" translation: "https://xinjignxiao.com/zh/articles/%E7%99%BE%E5%BA%A6%E5%A4%A7%E8%84%91%E4%B8%8E%E5%BF%AB%E6%B6%88%E4%B8%9A%E7%9A%84%E7%8E%B0%E5%9C%A8%E5%92%8C%E6%9C%AA%E6%9D%A5-7d60d620.md" attribution: "New Distribution — https://xinjignxiao.com/en/articles/baidu-brain-and-the-present-and-future-of-the-fmcg-industry-7d60d620/" citation: "谢永康. “Baidu Brain and the Present and Future of the FMCG Industry.” New Distribution, 2019-09-09. https://xinjignxiao.com/en/articles/baidu-brain-and-the-present-and-future-of-the-fmcg-industry-7d60d620/" usage_policy: "https://xinjignxiao.com/ai-policy.txt" --- # Baidu Brain and the Present and Future of the FMCG Industry > On August 21, at the 2019 China FMCG Conference hosted by New Distribution, Xie Yongkang, Senior Technical Manager of Baidu AI Technology Ecosystem Department, introduced the overall architecture of Baidu Brain, Baidu's AI technology culmination, and shared the AI capabilities it currently provides for the FMCG industry, as well as its future plans. **Follow and star ↑↑ [New Distribution]** **See how many friends are following industry trends with you** On August 21, at the 2019 China FMCG Conference hosted by New Distribution, Xie Yongkang, Senior Technical Manager of Baidu AI Technology Ecosystem Department, introduced the overall architecture of Baidu Brain, the culmination of Baidu's AI technology, and shared the AI capabilities it currently provides for the FMCG industry, as well as its future layout. **Overall Framework of Baidu Brain** Xie Yongkang first introduced the overall framework of Baidu Brain: **From bottom to top, Baidu Brain consists of the infrastructure layer, perception layer, cognition layer, platform layer, and AI security components that span all layers.** The infrastructure layer of Baidu Brain includes big data, deep learning frameworks, and large-scale computing clusters. The perception layer includes perception capabilities such as speech, vision, and AR. The cognition layer includes natural language processing and knowledge graphs, enabling machines to think and reason. All these capabilities are fully opened to the public through the Baidu AI Open Platform (ai.baidu.com). Currently, the Baidu AI Open Platform has provided over 210 AI capabilities cumulatively, with more than 1.3 million developers. Most of these capabilities can be integrated with business systems within a few hours. For customers with customization needs, Baidu Brain also provides development platforms such as EasyDL and the deep learning framework PaddlePaddle. **Baidu Brain's Technology Ecosystem in the FMCG Industry** Over the past two years, Baidu Brain has supported ecosystem partners in implementing intelligent business solutions such as store inspection, intelligent settlement, member recognition, traffic analysis, and document recognition, targeting business scenarios involving brand owners, distributors, and retailers in the FMCG and retail industries. Currently, the "offline channel intelligent verification" platform jointly built by Baidu Brain and ecosystem partner Ai Linggong for FMCG enterprises is available for consultation and purchase through the Baidu AI Market. With its mature gig worker system, Ai Linggong has accumulated over 1 million store service personnel, covering 830,000 hypermarkets and retail stores nationwide. By applying Baidu Brain's product recognition technology, the review efficiency of the Ai Linggong platform has been further improved, jointly providing brand owners with accurate and efficient offline channel intelligent verification solutions. **EasyDL Retail Edition: The First AI Model Training Platform Tailored for FMCG and Retail** After engaging with many FMCG and retail customers, Baidu Brain has been thinking about how to better align with the business needs of brand owners, distributors, and even retailers, and how to better promote the intelligent process of the FMCG industry, improve terminal execution efficiency, and reduce costs in various business links. In view of this, Baidu Brain, through various technical optimizations, released this year the first AI model training and service platform for product (SKU) detection specifically tailored for the FMCG industry: **EasyDL Retail Edition**. Technically, product recognition belongs to the category of computer vision object detection. However, compared with ordinary object recognition, product recognition has several differences. Let's analyze with a case of a certain beverage brand: First, ordinary object detection only requires capturing the target or object to form a certain amount of image data. However, products can present various angles on shelves, including looking up, looking down, and side views, which cause significant visual differences. Therefore, product detection and recognition require correspondingly labeling a large amount of data. As is well known, the cost of data collection and labeling is very high. Second, similar products from different brands often have very similar appearances and packaging. Ordinary object detection algorithms usually do not perform well in product detection. Finally, because shelves are relatively narrow and long spaces, it is often impossible to capture the entire shelf in one shot. Therefore, post-processing of recognition results is also more complex than ordinary object detection tasks. EasyDL Retail Edition has made deep technical optimizations to address these issues. First, for data collection and labeling, Baidu Brain and ecosystem partners have jointly built an SKU collection box. It supports simultaneous capture of panoramic images of products from multiple angles and seamlessly connects with cloud model training. For collected raw images, EasyDL Retail Edition uses data augmentation technology to combine images with shelves and cabinets, generating a large amount of labeled data that can be directly used for training, effectively solving the problem of requiring large amounts of training data. Then, EasyDL Retail Edition has also made in-depth optimizations to model performance. By introducing visual attention mechanisms, it improves the detection of fine-grained objects. **It adopts a detection + recognition segmented approach to handle ultra-large-scale product classification problems. By optimizing the NMS strategy in general object detection models, it solves the recognition problem of overlapping or occluded products.** At the same time, EasyDL Retail Edition is continuously improving to meet partner needs. At this conference, it also brought four major new features: **1. Dataset Management API:** For users who already have labeled product data, EasyDL Retail Edition supports uploading offline labeled training data directly to the training platform via API, maximizing the reuse of existing data and reducing the cost of model training. **2. Cloud Service Call Data Management Function:** Supports viewing images recognized by the product detection interface on the page, filtering out incorrectly recognized images, and after correcting labels, they can be used for iterative model optimization, continuously improving model accuracy. **3. Shelf Stitching Service:** For long shelves in real scenarios, EasyDL Retail Edition's shelf stitching service supports stitching multiple partial shelf images or videos into a complete shelf image and outputs product recognition results in the complete shelf image, meeting the business scenario of product recognition on long shelves. **4. H5 Experience Function:** After users train and publish a model, they can verify and demonstrate the model effect immediately without writing code. **Baidu Brain and the Future of the FMCG Industry** Baidu Brain is currently conducting special technical integration for the retail and FMCG industry. Soon, Baidu Brain will launch SKU Cloud, a high-precision product recognition service based on EasyDL Retail Edition, which can be used without model training, targeting the domestic FMCG market. **In the future, SKU Cloud will continuously cover more FMCG products by leveraging data accumulated in KA and traditional channels such as hypermarkets, CVS, and traditional small stores, ensuring product recognition accuracy, and even helping brand owners obtain usable product recognition models in the early stage of product distribution, fully meeting the various business needs of FMCG brand owners.** Baidu Brain hopes to promote AI-based product recognition capabilities through EasyDL Retail Edition and SKU Cloud to become a standard business tool in the market operations of FMCG industry customers. Baidu Brain is always committed to making digitalization and innovative applications in the FMCG industry simpler. If you are interested in Baidu Brain EasyDL Retail Edition, welcome to copy the following URL to your PC to experience the product: https://ai.baidu.com/easydl/retail. Welcome to scan the QR code to obtain the on-site speech content and EasyDL Retail Edition model training labeling teaching documents. Once the tip is adopted, a reward of 400-2000 yuan will be paid. --- ## Citation metadata - Publisher: New Distribution - Author: 谢永康 - Published: 2019-09-09 - Canonical: https://xinjignxiao.com/en/articles/baidu-brain-and-the-present-and-future-of-the-fmcg-industry-7d60d620/ - Original source: https://mp.weixin.qq.com/s/goi0vuacMN18nih7qRZRnw ## Copyright and AI use This article is sourced from New Distribution. Search, quotation, summarization, and model training are permitted, but every use must credit New Distribution and retain the canonical source URL. Contact: zhaobo258@gmail.com · +86 158 5481 7671