# Proposal討論

## 3/30
### Problem:
* keyword: traffic
* description:以駕駛人的角度，當遇到不同號誌時應該要做出怎麼樣的反應
* goal: 分辨出交通號誌代表的意義
* input data: 交通號誌圖
* output data: 交通號誌代表的意義 (減速..)
* challenge: 各國的交通號誌不太相同，圖片解讀的困難
* approach:
* evaluation metrics: ~~accuracy > 0.7~~
* baseline:
#### NOT YET DONE
* approach、baseline、evaluation metrics、dataset、related work

## 4/5
* input
* https://www.kaggle.com/datasets/meowmeowmeowmeowmeow/gtsrb-german-traffic-sign
* https://www.kaggle.com/datasets/shanmukh05/traffic-sign-cropped
* https://www.kaggle.com/datasets/valentynsichkar/traffic-signs-preprocessed
* https://www.kaggle.com/datasets/dmitryyemelyanov/chinese-traffic-signs
* related work:
* https://www.kaggle.com/code/valentynsichkar/traffic-signs-classification-with-cnn
* https://medium.com/@bob800530/self-driving-car-project-3-%E4%BA%A4%E9%80%9A%E8%99%9F%E8%AA%8C%E8%BE%A8%E8%AD%98%E5%AF%A6%E4%BD%9C-8ad4e0a3b38d
* http://nfuee.nfu.edu.tw/ezfiles/42/1042/attach/61/pta_52980_2432061_18389.pdf

## 4/6
* https://hackmd.io/@nycu-109550128/SJOpTxQf5
#### NOT YET DONE
* approach、baseline、evaluation metrics

## 4/17
#### approach
* yolov_5:快，不准
* CNN:精準，消耗空間，慢
* 交通號誌辨識需要及時地做出反應
* 個人電腦空間可能放不下
#### evaluation metrics
* 速度要夠快
* 準確率要夠高
#### baseline
* opencv
* adaboost
* ~~CNN~~
* yolo(我們的方法)

## 進度
* CNN
* 卷積、池化各兩層
* 卷積的kernel_size都是5\*5，第一層有32個filter，第二層64個filter
* 池化尺寸皆縮小一半
* 隱藏層的數量暫時設為512個
* 可修改參數:
* model.compile中的optimizer
* https://keras.io/optimizers/
* https://chtseng.wordpress.com/2017/10/01/%e4%bd%bf%e7%94%a8cnn%e8%ad%98%e5%88%a5%e8%be%a6%e5%85%ac%e5%8d%80%e7%8b%80%e6%85%8b-1/
* https://chtseng.wordpress.com/2017/09/23/%e5%ad%b8%e7%bf%92%e4%bd%bf%e7%94%a8keras%e5%bb%ba%e7%ab%8b%e5%8d%b7%e7%a9%8d%e7%a5%9e%e7%b6%93%e7%b6%b2%e8%b7%af/