ComputerVision

Everything about Computer Vision

For Beginners CNN in Action

• Tensorflow Playgroud
• Andrej Karpathy’s ConvNetJS

Objects (Inputs)

• Image
• Video

Types of Computer-Vision Tasks

• Classification
  • Binary Classification
  • Multiclass Classification
• Localization
• Detection
• Segmentation
  • Semantic Segmentation
  • Instance-based Segmentation

Model Zoo

• https://keras.io/api/applications/
• https://github.com/GoogleCloudPlatform/keras-idiomatic-programmer/tree/master/zoo
• https://paperswithcode.com/
• https://modelzoo.co/
• https://modeldepot.io/

Dataset	Description
ImageNet
Tensorflow Datasets

Model Architecture
Resnet

Approaches for training/inference

• Cloud Based API (inference)
• Pretrained Model (inference)
• Cloud Based Model Traning (traning/inference)
• Custom Training (training/inference)

  • Requirements

    • Dataset
    • Model Architecture
    • Framework
    • Hardware

Local Setup

• Lambda Stack
• ROCm - Setup of AMD GPU

Free Online Notebooks

• Colab
• Binder
• IBM Watson Studio
• Gradient

Cloud Visual Recognition APIs

• Microsoft Cognitive Services
• Google Cloud Vision
• Amazon Rekognition
• IBM Watson Visual Recognition
• Clarifai
• Algorithmia

Tools

• Tensorflow Datasets - for most optimize implementation

• Tensorboard - to visulize many aspects of training

• What-If Tool - to compare models

• tf-explain - Analyze decisions made by the network

• Keras Tuner - automatic tuning of hyperparameters in tf.keras

• For Any Question First Check this out

• Hyperas - Hyperparameter tunner

• Hyperopt - Hyperparameter tunner

• BayesianOptimization - Hyperparameter tunner

• AutoKeras - Automates Neural Architecture Search (NAS) across different tasks like image, text, and audio classification and image detection

• AdaNet - NAS

• AutoAugment - Utilizes reinforcement learning to improve the amount and diversity of data in an existing training dataset, thereby increasing accuracy

• TensorFlow Debugger - for debugging

• nvidia-smi - This command shows GPU statistics including utilization.

• TensorFlow Profiler + TensorBoard - This visualizes program execution interactively in a timeline within TensorBoard.

• OpenCV - Data Augmentation

• Pillow - Data Augmentation

• Knock Knock - Get notified when your training ends with only two additional lines of code

• Fast Progress Bar - Simple and flexible progress bar for Jupyter Notebook and console

• Netron - Visualizer for neural network, deep learning and machine learning models

• NN-SVG - for making neural networks diagrams

• PlotNeuralNet - Latex code for making neural networks diagrams

• Fatkun Batch Download Image - Chrome Extension to Download images

Some Important Steps

• Class Activation Map
• Transfer Learning
• Fine Tuning
• Data Argumentation
• Most Confident/Least Confident/ Incorrect with high Confidence
• Feature Extraction (feature vectors or embeddings or bottleneck features)

Deployment Platforms

• Computer/Server
  • HTTP servers
    • Flask, Django
  • Hosted and managed cloud stacks
    • Google Cloud ML
    • Azure ML
    • Amazon Sage Maker
  • Manually managed serving libraries
    • TensorFlow Serving
      • Using Docker
    • NVIDIA TensorRT
  • Cloud AI orchestration frameworks
    • KubeFlow
• Website (Browser)
• Mobile
  • iOS
  • Android
• Edge Device
  • Raspberry Pi
  • Jetson

Applications

• Reverse Image Search Engine (Instance Retrieval)
  • Similarity Search
    • K-Nearest Neighboors
    • Approximate Nearest Neighbors
      • Annoy
      • FLANN
      • Faiss
      • NGT
      • NMSLIB
• Siamese Networks for One-Shot Face Verification

Sources

MOOCs

Udemy:

1. Deep Learning: Advanced Computer Vision (GANs, SSD, +More!) (2020)

- https://github.com/lazyprogrammer/machine_learning_examples/tree/master/cnn_class2

Books

1. Practical Deep Learning for Cloud, Mobile, and Edge
2. Mastering OpenCV 4 with Python
3. Learning OpenCV 4 Computer Vision with Python 3 - Third Edition
4. OpenCV 4 with Python Blueprints - Second Edition

Github

• For book 1: https://github.com/PracticalDL/Practical-Deep-Learning-Book
• For book 2: https://github.com/PacktPublishing/Mastering-OpenCV-4-with-Python
• For book 3: https://github.com/PacktPublishing/Learning-OpenCV-4-Computer-Vision-with-Python-Third-Edition
• For book 4: https://github.com/PacktPublishing/Python-Machine-Learning-Blueprints-Second-Edition

Useful Tools and Links

• https://projector.tensorflow.org/ - Embedding Projector
• http://ann-benchmarks.com/ - benchmarking environment for approximate nearest neighbor algorithms search
• https://www.slideshare.net/anirudhkoul/deep-learning-on-mobile-2019-practitioners-guide - Presentation on DL on Mobile 2019

To Do

1. Build a Web App that takes an image and outputs whether that image is horse or human
2. Build a Web App that takes an image and produces similar images