Fashion State of Art

 

1/ Using AI to generate fashion

After a bunch of experimentation I finally got DALL-E to work for video by combining it with a few other AI tools

See below for my workflow -#dalle2 #dalle #AIart #ArtificialIntelligence #digitalfashion #virtualfashion pic.twitter.com/x3zP3fIp4G

— Karen X. Cheng (@karenxcheng) August 30, 2022

Hopefully soon will get into mainstream adoption..DALL-E..text to image, image to styles, styles to realtime .

Keep Exploring!!! 

Balancing code - data - business

To generate value, one needs to solve real problems, to solve problems needs to know business execution at ground level, this just translates to people, data, information, and touchpoints. Taking some info from ground level, Adding the pieces with your ideas, and Building a working idea is the balance of feasibility vs solution analysis. We may not know everything 100% but learning things on the go/need basis and learning to make progress in the interest of time is key. The feeling of once you do two or three examples, This is all it is all about that becomes paramount :)

The ability to learn / self-manage / progress with directions / see the big picture is more important than remembering code syntax. I believe it's a balance of all, not just algos. How you know where to apply, and when to apply is much beyond kaggle and competitive tests :)

• Learning business
• Knowing Touchpoints 
• Data Exchange patterns - Methods of information exchange (Batched / Real-time)
• Tech tools landscape
• Prototype code up
• 30% business, 30% data, 30% code to get perspective

My Hiring Recommendations

• Hire someone who knows what business problem they solve
• How they design based on time constraints
• Tradeoff on Tech vs Time vs Design
• Thinking for scale/performance
• Need not be full stack but knows how to communicate and get things done
• Customer is more important than individuals EGO's

Keep Exploring!!

Video Editing Tips

 

Keep Exploring!!!

Algos vs Features

You will see a lot of #posts on #Datascience #alogrithms but very few posts on #features #data. 80% of core learning is #features and #data. Hope #features are also #shared more openly for benefit of the community and #learning from #experiements. 

Pay deep attention to features. For example, For an average shopper what all I can track about his purchase?

• How many average views to buy a product
• How many competitive products comparison
• How many average price comparisons
• How many discount comparisons
• Time of purchase (Weekday / Weekend)
• Festive Shopper
• Bought together supplementary items
• Weekend shopper
• Late night shopper
• Salary Day Shopper
• Discount Shopper

Many times it is a tradeoff between 10000 feet view vs 1000 feet ground view. Balance both #perspective vs #bigpicture. Nth level personalization / real-time personalization across browsers/devices like we see how google recommends in the next browser session. #perspectives #datascience #persppectives #learningmoments 

Keep Exploring!!!

Question the Team / Yourself to understand Value of Delivery / Interests ?

Taking inspiration from link 

1. Cult of speed in the first 3-4 weeks - Have something useful for customers :) in every sprint
2. Ability to recover quickly from failures, Acknowledge and move on, Transparent
3. Create a culture of 'disagree and commit', Bring in complete perspective 
4. Are you working efficiently? Self-critical?
5. How much time do you spend on low ROI activities? Spot and remove barriers
6. How much time do you spend trying to figure out what to work on?
7. Are people working on projects that play to their strengths?

Keep Thinking!!!

Tesla CVPR Talk - [CVPR'22 WAD] Keynote - Ashok Elluswamy, Tesla

Key Notes

• Tesla autopilot for lane management, safety, lane changes, exits, blind spots
• 100k cars have self-driving beta software (parking lots, city streets, highway)

• Algos onboard computer
• 8 Cameras - 36 frames / second

• 360 views around cars
• No other sensors outside this
• Segmentation to classify Drivable / Not

• Depth is also not enough / Difficult for occlusion detection
• (Capture)
• Occupancy Networks

• Consistent output space
• Compute efficiently

• 3D point occupied or not
• Fixed queries attend over image space feature
• Positional embedding in image space

• Moving / Non-Moving obstacle
• Produce both moving/static in same frame
  
  
  
  
• Occupancy produces both include in occupancy (Static + Dynamic)

• Nerf - Scene creation
• Extract descriptions from scene
• Build perspective outside the world

• Occupancy to Collision Avoidance
• Based on direction and velocity estimate collision avoidance

A Survey of Deep Learning Techniques for Autonomous Driving

Twelve Questions about Tesla Autopilot Technologies Exposed at 2021 AI Day

Keep Exploring!!!

Deep Networks - Tuning Tips - Segmentation - Observations

1. Try to overfit your network on much smaller data and for many epochs without augmenting first (Link)

2. Explore variations of the network (Link)

3. Change layers number and units number.

4. Change batch size.

5. Add dropout layer. (Link)

6. Verify that your code is bug free There's a saying among writers that "All writing is re-writing" -- that is, the greater part of writing is revising. 

7. Keep a Logbook of Experiments

8. Look at individual layers, Build a simpler model first (Link)

9. Use nn.BCEWithLogitsLoss as the criterion instead of nn.BCELoss and pass the raw logits to it by removing the sigmoid. (Link)

10. In colab plus GPU + HighRAM worked better

Even after 100+Epochs doesn't seem promising.

Github code is a good start but needs to relearn, Start again :)

Keep Learning!!!

Redis - Examples

 In an eCommerce site

• Preserve user cart information
• User browsing history
• User past purchases
• Automatically Expire sessions on timeout

Redis - Remote Dictionary Service

Redis - open-source key-value database server

Keep Experimenting!!!

NoSQL One Liners

During NoSQL Session, After the first half of sessions. Learners shared their comparison of RDBMS vs NoSQL

An RDBMS includes functions that maintain the security, accuracy, integrity and consistency of the data

RDBMS - consistent and availability

All tables are related to each other in RDBMS and  in NoSQL all tables are independent of each other

NoSQL - partition tolerance at the expense of availability

ACID (atomicity, consistency, isolation, durability) is a set of properties of RDBMS  for transaction 

RDBMS fixed schema and No SQL have dynamic schema (Schemaless)

In NoSQL, we can exclude the information which is not required 

When to Choose what?

• High reads consistent data - RDBMS
• High writes low reads - HBase, Cassandra
• Document-based storage (multiple key-value pairs, or key-array pairs, or even nested documents) - Mongodb, Couchdb
• Key-Value stores are similar to maps or dictionaries where data is addressed by a unique key - Redis

Keep Exploring!!!

Beauty - Markets - Landscape - Opportunities

Beauty - Markets - Landscape - Opportunities

Paper #1 - Growing a Beauty Business in 2022

• Merging with the Metaverse - Meanwhile, artificial intelligence (AI) and augmented reality (AR) beauty technology, provider 
• An NFT is a digital asset that represents real-world objects like art, music, in-game items and videos
• Consumers will want their digital persona to reflect their identity and individual sense of style
• Consumers today want transparency. They want safe and high-performing products that are not harmful to the environment

Paper #2 - COSMETICS AND BEAUTY PRODUCTS in Mexico

• There are three segments within the beauty and personal care sector: the mass segment, the prestige segment and the premium segment. 
• Mass beauty products represent the highest share (74.05%), followed by premium products (13.94%) and prestige products (12.71%).
• Generally, European products are considered prestigious or premium, since they cannot compete in price with those produced in Mexico, and are valued for their brand image, quality and medium-high price

Trending

• Anti-aging products: These are on the rise not only among the older population, but also among younger people.
• Male hair care products. Men, who are more loyal customers than women, are playing an increasingly important role, and especially in Mexico they are in the habit of using gum or fixative.
• Ethical, healthy and environmentally friendly products. Also vegan products and with recyclable or biodegradable packaging. Companies must offer ecological and natural solutions. For example, solid formats instead of liquids.
• Skin and feeding. According to Mintel, the development of new products based on "super-ingredients" such as algae, kale, etc., has tripled in the last 5 years.
• Personalized beauty: Adaptation of the product to the particular characteristics or tastes of each client.
• E-commerce is still under developed, but gained importance during the pandemic, in addition to the cost reduction and flexibility it offers. 

Paper #3 - GLOBAL BEAUTY INDUSTRY TRENDS IN THE 21st CENTURY 

• The time-saving products are a response to the needs of today's ever-busy consumers who want to limit the amount of time spent on their daily beauty routine
• Long-lasting cosmetics are perceived to have a good price-value ratio, as they can be used more sporadically than the conventional products. Examples include: long-lasting lipsticks, nail polishes and 24-hour moisturizers in skincare
• What is becoming more and more visible in the global market are organic beauty products – those natural cosmetics, manufactured in accordance to the fair-trade philosophy. 

Paper #4 - THE FUTURE OF BEAUTY

• The secret is finding a connection between the micro-trend and macro consumer needs, which are readily shifting
• NATURAL - As with organic food and household supplies, more consumers are becoming interested in natural beauty products
• PERSONAL - The same shoppers who are redefining what healthier, natural beauty products look like are themselves younger, more racially, culturally and ethnically diverse and cosmopolitan consumers
• CONNECTED - As more buyers grow increasingly comfortable with incorporating online channels in their paths to purchase, the opportunity for manufacturers and retailers to connect with them is increasing as well.

Paper #5 - 8 Trends impacting the beauty industry

• The demand for more personalized products and services will continue which means companies must find a delicate balance between standardization and personalization.
• Consumers would like to make a more informed product decision and so enabling this can mean more brand loyalty.
• Fitness and health are prioritized
• Social media has become the new door-to-door for the global beauty industry with more engagement online than ever seen before. Consumer reviews will increase in importance and so greater focus needs to go on the way in which brands facilitate this community feedback and approach.
• Enhancing consumer convenience such as automated re-ordering based on usage may be an opportunity in the future.
• To increase loyalty, brands need to give consumers the ability to feedback and influence and this therefore could mean they become more involved in the product design process
• Product ingredients and where these were obtained will need to be made more available to consumers
• Consumers place greater emphasis on quality, environmental impact, ethics and sustainability.
• Use the data gathered to understand shopper interests e.g. bargain-hunters, vegan enthusiasts etc. and design products based on this.

Paper #6 - 2022’s Beauty Trend Forecast Report

• Retro Hair Revival It’s not just workplaces that are opting for a hybrid model; the infamous ‘70s mullet is making a comeback
• Body shaping continues to offer variety for those looking for a new silhouette. Traditional body contour continues to trend in popularity (up 35%)
• Nail art (up 75%) continues to trend with designs that are more intricate than ever
• Searches for hair Botox were up 22%.

Paper #7 - BEAUTY’S BRIGHTEST TRENDS

• Online or in-store? Beauty retailers of the future will be adapting to the new shopping habits of consumers
• Brick-and-mortar Rebound - consumers will once again be visiting stores to play catchup on the time they were stuck at home.
• Off-Mall Retailers - offering a more curated assortment of products from a variety of categories
• E-commerce to Stay the Leading Channel - Stores are becoming more digitally oriented with a wider adoption of assisted virtual selling tools
• In-person professional aesthetic treatments are making a strong comeback

Paper #8 - The Five 2022 Beauty Trends Personal Care Manufacturers Need to Know

• Trend #1: Consumer demand for ingredient transparency
• Trend #2: Strong appetite for sustainability
• Trend #3: Rising demand for personalization
• At the cutting edge of the industry, personal care brands are using AI and machine learning capabilities to take customer insights to a new level,
• Delivering fully personalized products using valuable information being generated by every customer interaction. 
• For instance, skincare brand Atolla is continually adapting its formulations using deep learning technology, while L’Oreal’s Perso at-home skincare and cosmetics device harnesses AI to deliver personal analysis and environmental assessment, combined with customizations to product preference and dispensing methods. 

Paper #9 - BEAUTY T R E N D S R E P O R T 2022

• With the increase in skincare knowledge comes a focus on skinhealth, not just skin-care, as the way to great skin for life

Best-selling

• Facial Skincare brands with the biggest growth in sales at Boots this year are science-based brands
• The rise in living costs doesn’t mean we have to revert to buying throwaway cosmetics. Affordable Beauty doesn’t have to mean compromising on quality or our values either.

Paper #10 - 2022 State of Beauty Industry

Paper #11 - TRENDS & INSIGHTS

Keep Exploring!!!

Experience - Exposure - Domain - Data - Diversity Thinking

• My First year - Windows98 Testing
• Second year - C/C++ Mq Adapter Design
• Third year - Application support / migration / ITSM / Deployment / Production Support
• Fourth Year - SQL Migration / Performance Testing
• Fifth Year - Biztalk / SQL Migration / Replication / Developer
• Sixth Year - OLTP SQL Developer 
• Seventh Year - BI + OLTP Developer / Warranty migration 220 million records (Supply chain 4th year :))
• Eighth Year - QA Manager / Prod support / On-call / People Management
• 9th Year - DB / QA Developer / Product Transition
• 10th Year - Forecasting Feature in Product / DB / QA / Automation
• 11th Year - Hardware Integration / Professional Services / SQL Developer
• 12th Year - AWS Scaling / Perf testing on horizontal scaling for Tag counting / Function + Perf / DB
• 13th Year - SQL 2016 Migration Analysis / Big Data Architecture Analysis
• 14th Year - AI / ML - PG / DB Dev / BI
• 15th Year - AI / ML - PG second year / DB Dev / BI / Freelancing
• 16th - AI Freelancing / Training / DB / BI Dev / Patents
• 17th  -  AI Freelancing / Training / DB / BI Dev / Patents
• 18th - Vision / Forecasting / Recommendations  / People Management / Training
• 19th - Vision / Forecasting / Recommendations  / People Management / Training

All the different parts of experiences sum up and help me in knowing the Data journey / AI journey/business challenges 

It was varied roles / multiple domains and products but everything was worth it :)

Thanks to all my previous companies/training providers / past employers/freelance offered by startups...

Multiple lenses - Development - QA - Automation - Performance - Support - BI - Database performance tuning - Computer vision - Forecasting - Recommendations - Many times I have influenced / implemented product key features based on domain expertise.

• Warranty Migration of 220 million SKUs
• 3PL touch point integration
• Spot capabilities in Product with Data & AI (What can we do with what we have :) 

Keep Exploring!!!!

Most impactful keywords in your Search - Fashion & Beauty

Good ranking for these keywords will have good conversions :)

Ref - Link

Keep Exploring!!!

Image catalog management ? Opportunities ? Fashion & Beauty

Do you know Image types that work best for beauty and fashion ?

• Outdoor fashion images have 195% more conversion rates than indoor images
• A person with a beauty product has 27% more conversion than just image of product
• Image with no text has 31% more conversion than image with text
• Distant images have 106% more CVR than close up images

Opportunity

• For fashion provide a template for outdoor image
• Remove watermark from images
• Provide Far view / Validate images by near/far view
• Categorize product catalog with people with products vs just product images

Ref - Report

Keep Exploring Vision Opportunities!!!

Image Quality Reads

Ideas for Image Features and Image Quality

• Dullness - analysis of prominent colors 
• Image Whiteness - color analysis
• Uniform Images (with no pixel variations)
• Dominant Color

Key Topics to Explore

• Super-Resolution
• Image Deraining
• Image Dehazing
• Deblurring
• Denoising
• Image Restoration
• Image Enhancement
• Image Demoireing
• Inpainting
• Image Quality Assessment

Ref - Link

Projects to Explore

• GIQA: Generated Image Quality Assessment
• Deep Image Structure and Texture Similarity (DISTS) Metric
• Image Quality Assessment
• Image quality is an open source software library for Automatic Image Quality Assessment (IQA)

Keep Exploring!!!

OCR again - 2022 Updates

Donut 🍩, Document understanding transformer, is a new method of document understanding that utilizes an OCR

• Based on the transformer concept
• Experimented with the sample colab code
• Gradio is like streamlit 

Samples and Results

Demo codes Link

Keep Exploring!!!

Summary Fashion Attributes

Clothing Recommender System

• Part I : Object Detection
• Part II : Attribute Tagging
• Part 3 : Recommendation based on Frequency

Fashion Meets Computer Vision: A Survey

Paper #1 - Progressive Fashion Attribute Extraction

• Attributes (neck design detailing, sleeves detailing, etc) 

Paper #2 - Attr2Style: A Transfer Learning Approach for Inferring Fashion Styles via Apparel Attributes

• Low-level attributes of an apparel (for example, neck type, dress length, collar type, print etc)

• Transfer learning based approach to address the issue of style-based image captioning for a target dataset

Paper #3 - The iMaterialist Fashion Attribute Dataset

Paper #4 - A Deep-Learning-Based Fashion Attributes Detection Model

Paper #5 - FashionSearchNet-v2: Learning Attribute Representations with Localization for Image Retrieval with Attribute Manipulation

Myntra Customization

Occasion based Recommendation system in E-commerce like Amazon, Etsy

Visual Attributes for Fashion Analytics

We use low-level visual features to predict intermediate clothing attributes such as color, pattern, material, or collar type Occasion-oriented clothing recommendation

Attribute Types

Color/ Attributes

More Reads

Fine-Grained Fashion Similarity Prediction by Attribute-Specific Embedding Learning

Keep Exploring!!!