TuDo1403 / DS102.L11.CNCL

Object Detection

DS102.L11.CNCL Object Detection

A. Huong dan cai dat

I. Setup environment

1. Cai dat Linux Terminal tren Windows

• Link: Microsoft
• Tai Ubuntu 20.04 tren Windows Store

2. Update terminal

• sudo apt update

• sudo apt upgrade

• sudo apt install gcc

3. Cai dat anaconda tren ubuntu terminal

• Tai anaconda cho ubuntu (file dang .sh)

• chmod +x ten_file_anaconda.sh

• ./ten_file_anaconda.sh

• use conda init -> yes
• Tat terminal hien tai roi mo lai

• conda update --all

4. Tao environment de lam viec

• conda create --name ds102 tensorflow-gpu

• conda activate ds102

5. Cai package de lam viec

• conda install matplotlib

• conda install pandas

• pip install pickle5

• pip install torch torchvision

• conda install scikit-learn

• pip install opencv-python

II. Cai dat Latex

• Mo terminal

• sudo apt install texlive

• sudo apt install latexmk

• Neu bi loi thi dung lenh sau search package bi thieu

• sudo apt-cache search "package"

• Cai package

• sudo apt install "package"

III. Set up VSCode

1. Download + install vscode

• Link: VSCode

2. Cai dat extension

• LaTex Workshop
• Live Share
• Remote - WSL

B. Noi dung can tim hieu

I. Image Processing

1. Normalizing (Range 0-1)

2. Random Cropping

3. Random Flipping

4. Cut out regularization

5. Image Augmentation (Optional)

II. Hyperparameter Tuning

1. Grid Search

2. Random Search

3. Evolutionary Search

4. Bisection

III. Evaluation Metrics

1. K-Fold Stratified Cross-validation

2. IoU

3. AP

IV. Machine-learning Models for Object Detection

1. Random Forest

2. SVM

V. SOTA Deep-learning Models

1. YOLO Algorithm

2. R-CNN

3. Sliding Window

C. Road Map

Stage I

• Tien xu ly anh
• Code xong phan log du lieu moi experimental run
• Code xong phan luu du model moi experimental run
• Code chay ko bug
• Viet report 25%

Stage II

• Chay xong cac thi nghiem
• Plot cac figure danh gia
• Build cac table de danh gia
• Viet report 75%

Stage III

• Lam slide seminar
• Viet report 100%
• Code demo ket qua