# Image Recognition - pokemon

[TOC]

## Files and Folders

.
├── app.py
├── dataset10
│   ├── Arbok
│   ├── Dragonair
│   ├── Electrode
│   ├── Golduck
│   ├── Golem
│   ├── Haunter
│   ├── Mew
│   ├── Pidgeot
│   ├── Poliwhirl
│   └── Rattata
├── labels.txt
├── my_resnet
└── pygamee.py

## Motivation

- programming
- take a peek at AI

> programming - giving a set of instructions to a computer to execute
> AI - Artificial Intelligence (Machine Learning)

## Keys to archive GOAL

- basic knowledge about **programming**
- basic knowledge about **machine learning**
- how does computer deals with **images**

## Preparation

- programming language - python
- [why python?](https://github.blog/2023-03-02-why-python-keeps-growing-explained/#using-python-for-and-artificial-intelligence)
- laptop
- webcam

## General Steps

1. gather **data** (images)
2. train the **model** to recognize
3. validate model's **performance**

## Dive Into All Tasks

### gather data

- required a lots of images
- to learn to recognize these images, computer needs to see as much as possible
- take photos on our own
- use datasets that someone collected (open source)
- goods and bads
- precise
- quantity
- imbalanced
- only sample few kinds of pokemon in our task
- 15

### train the model

machine **learning** is basically math and probability, but it's hard to implement all these equations and theories on our own.

machine learning **framework** comes in very handly. choose to use [keras](https://keras.io/).

with general ideas about our task, we can build powerful AI too.

> machine learning - computer learn to remember something without being explicitly programmed to do so.

### model performance

> model validation - to ensure that our ML/AI model is performing as it should.

- accuracy
- the higher the better
- downside of accuracy?
- loss
- the lower the better

*to-do: display model performance (accuracy, loss)*

## Problem and Solving

1.

## More and Beyond