Nicholas Renotte - Tensorflow Object Detection Walkthrough (Anaconda Version) - 2

Reference

https://www.bilibili.com/video/BV1Aq4y1E75N
https://github.com/nicknochnack/TFODCourse
https://github.com/tzutalin/labelImg

Edible Instruction

1. Original video is using python venv, it didn’t work with my python3.9, so I removed 3.9 and installed 3.8
2. Although many people said Anaconda sucks, but it’s so easy(GUI) to create new profile, and Jupyter Notebook also included. (Only need one command to link the profile)
3. I’m also a novice of AI, if you can’t comprehend those things, just keep trying. you’ll be success one day!

Error Guide

The Jupyter notebook didn’t work after PC reboot.

Error message as down blow (Not mine),

Solution: (refer to https://www.jianshu.com/p/b3f076182fe3)

pip uninstall pyzmq
pip install pyzmq
(execute it on tfod profile)

Can’t install --upgrade pyqt5 lxml on Jupyter notebook

Solution: Pyqt5 is too big, and need run install many times.

Collect images

1. Git clone https://github.com/nicknochnack/TFODCourse

2. open 1. Image Collection.ipynb with Jupyter notebook
   

3. Press Shift + Enter for each cells

   Note: cap = cv2.VideoCapture(0) you may need to change the number to find your own camera.

4. 5 pictures for each gesture.
Make sure you got the clearly pictures. if not, do it again.

Image Labelling

1. pip install --upgrade pyqt5 lxml

2. Create folder and git clone another repositories
   the command didn’t work on my Jupyter, so I git clone manually.
   

3. Open labelimg application
   

4. Label your images as video tutorial.

5. Create and copy files to train and test folder