Use Python Tensorflow for machine learnning
Machine learning using libraries such as tensorflow is about as fashionable as it gets in the computer science industry at the moment. It is a rapidly moving field and finding good tutorials and resources that are (a) current and (b) pitched for beginner programmers can be a challenge. Of the several I have received, this looks to be quite good - although you should only embark on something like this if you have several projects worth of experience already.
The written version of the above tutorial is on his github page. I recommend skim reading the github page as you watch the video.
Download and install these first:
- How to Install TensorFlow on Windows: https://youtu.be/RplXYjxgZbw
- Download and install CUDA v9.0 and cuDNN v7.0
- Install Anaconda with Python 3.6 as instructed in the video
- TensorFlow Object Detection API repository: https://github.com/tensorflow/models
- TensorFlow Model Zoo page: https://github.com/tensorflow/models/...
- LabelImg utility: https://github.com/tzutalin/labelImg#...
Time codes in the video:
- 1:54 Step 1. Install TensorFlow-GPU
- 3:14 Step 2. Set up Object Detection directory and Anaconda virtual environment
- 15:21 Step 3. Gather and label pictures
- 18:35 Step 4. Generate training data
- 20:16 Step 5. Create label map and configure training
- 23:46 Step 6. Train object detector
- 26:54 Step 7. Export inference graph
- 27:45 Step 8. Try out your object detector!!
It might be worth checking the relevant reddit post as there are a couple of corrections to the tutorial (though mostly good reviews as well):