快速搭建 Milvus 以图搜图 v2.0

1 .Milvus image search system v2.0 

2 .Agenda • Milvus 以图搜图 1.0 • Milvus 以图搜图 2.0 原理 • Milvus 以图搜图 2.0 demo 演示 • Q&A 

3 .Zilliz intro 

4 .Zilliz: Who are we? • Open-source software company based in Shanghai • Mission: Reinvent data science • Main contributor of Milvus project 

5 .We are hiring~ Find our open positions: • Backend developer (C++/Golang) • AI algorithm engineer (CV/NLP) • Frontend developer • Product manager • Cloud infrastructure engineer • Technical Writer • And a lot more… You may also contact TA@zilliz.com 

6 .Milvus project intro 

7 .Unlock the treasure of unstructured data AI algorithms transform image, video, voice, natural language into vectors, and enables understanding and utilization of unstructured data at scale. Unstructured data Deep learning models Embedding vectors Knowledge, insight, $ 

8 .Milvus Big Picture Inference Layer Data Service Layer 

9 .PaddleRec x Milvus Tutorials: Baidu AI Studio- https://aistudio.baidu.com/aistudi o/projectdetail/1816335?channelT ype=0&channel=0 PaddleRec Github- https://github.com/PaddlePaddle/ PaddleRec 

10 .Milvus: The journey 2018.10 2019.04 2019.06 Milvus 1st seed The idea 0.1 user Joined Milvus Open LF AI & 1.0 Source Data 2019.10 2020.03 2021.03 

11 .Progress 6.8K 6.1K 136 Commits GitHub stars Contributors 20 1000+ 19 Release Users Patents filed 

12 .Useful Links Performance benchmark: https://milvus.io/docs/benchmarks_aws https://milvus.io Live demo: https://github.com/milvus-io/milvus https://www.zilliz.com/solutions https://www.milvus.io/scenarios https://milvusio.slack.com https://twitter.com/milvusio • Content-based image retrieval • QA chatbot powered by NLP https://medium.com/unstructured- • Molecular analysis data-service https://zhuanlan.zhihu.com/ai-search WeChat Public Account: 

13 .Meet our speaker 

14 .Speaker bio Data Engineer •Data preprocessing •AI model application •Nginx & php •Docker deployment 

15 .Milvus Image search system v2.0 p Background p How to use p Practice 

16 .01 Background 

17 .v1.0 l The core of the architecture is the VGG model, which converts images into vectors, and Milvus, which stores the vectors and performs similar vector retrieval. 

18 .Live demo https://zilliz.com/solutions https://milvus.io/cn/scenarios 

19 .v2.0 01 Milvus 1.0 Milvus has been upgraded from version 0.10.0 to Milvus 1.0. 02 YOLOv3 Model A new function for image target detection has been added, here using the yolov3 enhanced model, which is relatively more accurate and faster. 03 Resnet50 Model Compared to the VGG model, the Resnet50 model is not only faster to train, but also has a higher accuracy in image recognition. 

20 .02 How to use 

21 .Environment requirements The following tables show recommended configurations for reverse image search. These configurations haven been tested. Component Recommended Configuration CPU Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz Memory 32GB OS Ubuntu 18.04 Milvus 1.1 Software pic_search_webclient 0.2.0 

22 .Dataset The image dataset used in this test is the PASCAL VOC image set, containing 17125 images covering 20 catalogs: human; animals (birds, cats, cows, dogs, horses, sheep); vehicles (airplanes, bicycles, boats, buses, cars, motorcycles, trains); interiors (bottles, chairs, dining tables, potted plants, sofas, TVs). Data set size: 2GB Download location: http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar 

23 .How to use Github Link : https://github.com/milvus-io/bootcamp/blob/master/solutions/image_search2/README.md Ø Run Milvus Docker Milvus 1.1 Install Link:https://milvus.io/cn/docs/v1.1.0/milvus_docker-gpu.md Ø Install the Python packages you need $ cd /image_search/webserver $ pip install -r requirements.txt 

24 .How to use Ø Modify configuration file Parameter Description Default setting MILVUS_HOST milvus service ip 127.0.0.1 MILVUS_PORT milvus service port 19512 VECTOR_DIMENSION Dimensionality of the vectors 2048 DATA_PATH Path to save images /data/jpegimages DEFAULT_TABLE The milvus default collection milvus_183 UPLOAD_PATH Storage path for uploaded images /tmp/search-images /yolov3_detector/data/yolov3_darkne COCO_MODEL_PATH Path of the target detection model t /yolov3_detector/data/yolov3_darkne YOLO_CONFIG_PATH Profile path for target detection model t/yolo.yml 

25 .How to use Ø Start query service $ cd /image_search/webserver/src $ python app.py Ø Run pic-search-webclient docker $ docker run --name zilliz_search_images_demo_web -d --rm -p 8001:80 \ -e API_URL=http://${WEBSERVER_IP}:5000 \ milvusbootcamp/pic-search-webclient:0.2.0 

26 .03 Practice 

27 .Outlook l Continuous optimization of the revise image search system l Plan to provide revise image search API 

28 .Q&A 前10位问问题的小伙伴将可以获得限量的 Milvus 卫衣 （左图👈 ）领取奖品请加小助手微信：zilliz-tech，或是 扫下面的二维码。👇 

29 .Thanks