Visual Search as a Cloud Service by Large-Scale Commodity GPU Adoption
Ashwin Nanjappa
Visenze
Outline
Outline
Who we are
Our background
What we do
What we do
One-click visual search experience
object detection + visual search
60+% faster than text search
ViSenze visual search API helps people find visual information more easily
Lazada
Goodrich
Patsnap
ViDiscovery/ViSearch App
What we do
- 75% manual effort
ViSenze’s auto tagging helps sites tag their products more efficiently
ViSenze’s auto tagging solution helps sites tag their products more efficiently
Deep structured visual taxonomy for specific verticals
Deep structured visual taxonomy for specific verticals
Video Recognition API service
Application: Video + shopping user experience
Technologies
Computer vision and Deep learning
Distributed web service development
Visual search infrastructure
Visual search infrastructure
Visual search infrastructure
Visual search infrastructure
Offline training
Offline training: Siamese network
A loss function more suitable for visual search
Offline training infrastructure
Experiment as a service
Experiment infrastructure
Online search with feature and recognition
Online infrastructure
Online infrastructure
Search
Index
Online infrastructure: indexing
API load balancers, index queues, customer queues
Online infrastructure: indexing
AWS Cx instances, 200 max, throughput: 1M images/hour
Online infrastructure: indexing
Features on S3, hashes loaded into memory on region servers
Online infrastructure: search
Load balancers and API servers for low latency
Online infrastructure: search
AWS G2 instances to detect and extract features
Online infrastructure: search
Distance search using hash and features
Online infrastructure: search
Latency: <400ms, Bottleneck: Getting the image!
Amazon GPU instances
Amazon GPU instances
Not available in all regions (US only)
Hybrid infrastructure
Everything on cloud GPU: not there yet
GPUs @ Visenze
Thank you!