Deep SQL

 What is in the world of SQL

SQL is Deep, Know your depth of competency :)

Keep Exploring!!!

Building Vision Models - Myth vs Reality

Challenge / Perception - Customer feels we need millions of images 

Reality - We do not need to wait for perfect data, Data collection, or Synthetic data creation everything is a going process

Challenge / Perception  - Data collection is effortless, It can be done by google search / kaggle

Reality - The real world and the kaggle dataset are miles apart. Real-world challenges are dependent on light/angle/hardware used. Buying data is even more expensive :). Data cost is more costly than model training time

Challenge / Perception  - I need the start of the art model with 99% accuracy / Can we get a performance like the state of the art? 

Reality - We need to be realistic with the data we have, and an incremental model that we can develop. 

Challenge / Perception - Model development is a one-time effort. Collect / Build / Deploy / Move on 

Reality - Base model / retrain / field test and next version is incremental effort. ML is an iterative incremental effort. It has a set of parallel ongoing efforts like below

When the customer wants state of art but has no strategy on how they need to incrementally build upon becomes an effective challenge to provide the vision/clarity.

The building is easy sometimes vs Selling is hard many times. 

Keep Thinking!!!

Retail Product Detection

Ref - Post 

• Product region, brand logo region
• Product textual data (title, brands)
• The regions of interest in images were detected by a pretrained teacher model
• Following the trend of using free-form text, we train the CPG model with 2.3M product entities synthesized from an e-commerce site in a self-supervised fashion
• The bounding boxes for product-noun-to-object task are generated by a pre-trained general domain modulated detection model
• Visual-language understanding of logos, brand strings, product details for the query product entity and for all brand representative product entities

• Text to image lookup and comparison
• Similar embedding lookup and comparison

• Crafted image caption is tokenized and encoded using a pre-trained text encoder: RoBERTa
• Image and textual features are concatenated as a multimodal vector and fed to a joint transformer encoder with cross attention between image and textual features

Keep Exploring!!!

Vision and Manufacturing

 

Keep Exploring!!!

Vision and Retail / LEGO

 

• Manual process to automated process
• High-quality image checks
• Better attention-based models

Ref - Link

Keep Exploring!!!

Computer Vision Landscape 2022 Report

Why Computer Vision?

Hardware used

• High-resolution types
• Compact types
• High-speed types
• Standard types

Illumination techniques

• Backlighting - Detect the presence or absence of gaps or holes, identify bubbles, cracks, or scratches
• Dark field lighting - Features like edges, scratches, or notches
• Bar lighting
• Dome lighting

Key Works

Change Detection Techniques

• Frame differencing.
• Background subtraction.
• Motion segmentation.

Tracking Techniques

• ByteTrack
• Simple Online And Realtime Tracking (SORT)
• DeepSORT
• TransMOT
• FairMOT

From Post

Keep Exploring!!!

AI Vision - Energy Domain

 

• Vison for Windturbine inspection

• Types of Defects / Vision Implementation
• Energy Assets Monitoring

• Different damages - Erosion / Cracking
• Remote inspection with drones

• Thousands of photographs
• High Res pictures
• 4 hours for inspection for single turbine

• Identify different types of defects
• Plot the location

Keep Exploring!!!

AI Vision - Shell

 

Eyes - See, Brains to Think

• Remote inspections
• Warehouse / Counting Stock
• Robots / Drones for Refinery Management
• Safety / Monitoring Traffic
• Robots / Drones / Cameras / Handheld devices - Data Collection

• Safety and Compliance
• Reduce Exposure to field

• Spilling
• Analog Meter Reading
• Track Active Lights

• Follow map
• Collect data
• Analyze and send

• Frames - Detect Clock Region - Classify Clock / Non-Clock / Locate region / Compute angle of the needle
• Value - open/closed
• Needle  - Angle
• DL - Vision Analytics

• Adoption Recommendation
• Phased Adoption of Video Analytics
• Data Source - Model Building - Prototype
• Image Anamoly Detection - Darker image same type of day
• Variations of normal

Ref  - Link

Ref - Link

Keep Exploring!!!

Avatar = Magic of Computer Vision

• Pose Estimation
• Pose Detection
• Expression Capture
• GAN to reconstruct

State of Art = Vision in Action

Keep Exploring!!!

Production Detection from logos

Ref from post, post1

• Detecting all the possible image regions that might contain a logo
• Matching the detected regions against an ever-evolving set of logo prototypes
• Representation of an image with its text description
• Zero-shot logo recognition relies heavily on an embedding model for matching query regions against a constantly evolving set of cropped logo images
• Text pipeline to extract the text in the image via optical character recognition (OCR)
• We explicitly pair positive examples with their hard negatives, to encourage the model to distinguish logos from similar texts
• Create embedding for similar products and assign appropriate weights

Keep Exploring!!!