"No one is harder on a talented person than the person themselves" - Linda Wilkinson ; "Trust your guts and don't follow the herd" ; "Validate direction not destination" ;

December 24, 2022

Model Training Checklist

Often I end up running a working code but poor segmentation results. Need to save cost / try low res to slowly make changes.

Making a working model on a limited set is key.

  • Run for limited set 1000 images
  • Run for smaller resolution 128x128
  • Run for smaller batches
  • Model Compilation
  • Batch Compilation
  • Adjust learning rates

Keep Exploring!!!

December 22, 2022

Sagemaker - Notes

Key Notes

  • Create endpoint
  • Expose as Rest API
  • Realtime / Batch inference
  • Async for image
  • Deploy on CPU / GPU 
Factors for Deployment

  • Model Complexity
  • Payload
  • Complex Workflow
  • Compute, Storage, Networking Cost
Large Language Models Supported

  • Parallelize the model
  • Latency of low millisecond
Async  / Vision
  • Input uploaded in the bucket
  • Output loaded in bucket



Lower cost and higher throughput


Cold Start Latency 



Ops Features





Keep Exploring!!!

December 21, 2022

Deep Learning Markets - Use cases

 Combination of Vision, Forecast, Recommendation, Anamoly Detection, Optimization


Ref - Link

Keep Exploring!!!

Virtual Try on - Products / Demos

 





 




  


Ref Link
Slides Link 

Keep Exploring!!!

December 19, 2022

Vision In Manufacturing

 Another interesting use cases from link

Key Features

  • Targeted for visual inspection tasks in manufacturing environments
  • Docker container
  • Segmentation and localization
  • Active learning - identifies suspect examples quickly for human review and labeling
  • Auto-align images from camera streams


Visual Auto Inspection - With human in loop will be robust :)

Keep Exploring!!!



SmartRetail - GCP Architecture

 


Have been experimenting/pitching with little or no success. Either people see it as a commodity or see it as an extensive hardware investment. The willingness to experiment with key sections / innovate is hardly there.
  • People counting
  • Smart Shelf
  • Restocking
  • Heatmaps
  • Dynamic Ads
  • Sweethearing
Some key slides from Talk


  • Vision for item placement
  • Vision for item counting
  • Vision for Ads placement







GCP Architecture


  • Models run on GCP Kubernetes Engine
  • API Gateway
  • Images stored in Blobs
  • Results in Postgres / BigQuery


  • Apps for Endusers
  • Notifications
I have pitched and failed but happy to see Infliect, Tangoeye, and other companies shining in this space.

Keep Exploring!!!

December 15, 2022

Food - Recipe - Research Reads

Food Recipe Recommendation Based on Ingredients Detection Using Deep Learning

  • Custom dataset consisting of 9856 images belonging to 32 different food ingredients classes
  • Convolution Neural Network (CNN) model was used to identify food ingredients, and for recipe recommendations
  • Open Computer Vision Library (OpenCV) [2], TensorFlow [3], NumPy [4], and Keras [5]


Food Ingredients Recognition through Multi-label Learning


Nutrition5k: A Comprehensive Nutrition Dataset

Attention networks can extract much richer descriptions from the images compared to pure convolutional networks

Mining Discriminative Food Regions for Accurate Food Recognition

Taking inspiration from Adversarial Erasing, a strategy that progressively discovers discriminative object regions for weakly supervised semantic segmentation

The proposed architecture denoted as PAR-Net is end-to-end trainable, and highlights discriminative regions in an online fashion

On three food datasets chosen (Food-101, Vireo-172, and Sushi-50)

A Comprehensive Survey of Image-Based Food Recognition and Volume Estimation Methods for Dietary Assessment

Firstly, the visual representations of food images are of fundamental importance as it significantly impacts classification performance 

Therefore, many food recognition methods employ handcrafted features such as shape, colour, texture, local

As consolidated large food image datasets, for example, UECFOOD-100, Food-101, UECFOOD-256, UNCIT-FD1200, UNCIT-FD889






Deep Cooking: Predicting Relative Food Ingredient Amounts from Images

One method of predicting the ingredients given a food image is cross-modal recipe retrieval which outputs the ingredients and the corresponding amounts of the retrieved recipe

We use a Resnet50 [10] pre-trained on UPMC [21] and replace the last layer with ingredient amount prediction.

Food Ingredients Recognition through Multi-label Learning

CuisineNet: Food Attributes Classification using Multi-scale Convolution Network

Summary of Techniques

  • Object Detection
  • Multi label classification
  • Sliding Window Detection
  • Crop / Zoom Detection
  • Handcrafted features / Edges / Contours
  • Attributes (Color / Shape / Text Extraction)

Datasets

  • http://123.57.42.89/FoodComputing-Dataset/ISIA-Food500.html
  • https://github.com/ustc-vim/vegfru
  • http://123.57.42.89/FoodProject.html
  • https://github.com/monajalal/Kenyan-Food

Keep Exploring!!!

Edge Devices Notes

Paper - DeepEdgeBench: Benchmarking Deep Neural Networks on Edge Devices*

  • Raspberry Pi
  • Nvidia’s Jetson Nano
  • Asus Tinker Edge R
  • Raspberry Pi 4
  • Google Coral Dev Board

Nvidia Jetson Nano:

  • Nvidia deploys their Graphics Processing Unit (GPU) modules to the Edge with accelerated AI performance
  • The Jetson Nano’s GPU is based on the Maxwell microarchitecture (GM20B) and comes with one streaming multiprocessor (SM) with 128 CUDA cores
  • Compared to the other target edge devices in this work, the Jetson Nano stands out with a fully utilizable GPU

Google Coral Dev Board

  • The Google Coral Dev Board is one of the offerings which features the “edge” version of the TPU (tensor processing unit)
  • To make use of the dedicated unit, models in a supported format can be converted to work with the PyCoral and Tensorflow Lite framework

DeepEdgeBench Testing Framework

Raspberry Pi 4:

  • 4 comes with Gigabit Ethernet, along with onboard wireless networking and Bluetooth.
  • 4th generation is available in multiple variants with different RAM sizes (2, 4 and 8GB)

Tensorflow Lite: TensorFlow Lite was developed to run machine learning models on microcontrollers and other Internet of Things (IoT) devices with only few kilobytes of memory

TensorRT: NVIDIA TensorRT [36] is a C++ library that facilitates high performance inferencing on NVIDIA graphics processing units (GPUs)

Tensorflow: TensorFlow [24] is an open-source software library for numerical computation using data flow graphs. Nodes in the graph represent mathematical operations, while the graph edges represent the multidimensional data arrays (tensors) that flow between them.


Test Scenarios

  • Average time spent on inference of one of 5,000 images
  • Total time spent on inference of 5,000 images
  • Power during idle state (LAN on and off)
  • Average power during inference of 1,000 x 5 images
  • Total power consumption during inference of 1,000 x 5 images
  • Accuracy for each platform-device combination

Extend the FFmpeg Framework to Analyze Media Content

ML with Intel OpenVINO Toolkit — Super-Resolution

Keep Exploring!!

December 12, 2022

1 - 1 Discussions / Signs to Monitor Proactive Workers

Some tips for connecting/providing feedback

  • Learn without being told
  • Focus on outcome
  • Help others advance in their ideas
  • Open to new ideas
  • End of the day we are here to make money, so Why not focus on business  
  • The bigger the challenges, you need to persist and focus on making things work 

Keep Thinking!!!

December 05, 2022

Self appraisal - Note to Self and Team

Always ensure you can improve upon, Your next version better than your current version

  • Predictability of tasks, How well our estimates and actuals
  • Completeness and quality speak for it
  • 80% thinking where all it fails, scalability, Design, and 20% ensuring you code all possible scenarios.

Keep Learning!!!