Code Example - KIMDAWUN/DOJumpShot GitHub Wiki

step 1) Retrain ( DOJumpShot/Retrain/inceptionv3_retrain.py )

- Number of images: 2800
- Training steps: 4000
- Learning rate: 0.01
- Testing percentage: 10% of Dataset
- Validation percentage: 10% of Dataset
- Train batch size: 100
- Test batch size: entire test set
- Validation batch size: 100


step 2) Optimization

1. Remove unnecessary nodes  ( DOJumpShot/Retrain/strip_unused.py )

remove node

2. Convert into TFLite

convert tflite



step 3) Integrating deep learning model into Android Studio

1. Upload the mobilenet_v2_2800.lite and label.txt created as a result of retrain and optimization.

2. Classifies a frame from the preview stream.

classifies a frame from the preview stream

3. If jump accuracy exceeds the value of threshold, take a picture automatically. (Threshold: 0.9)

threshold가 0 9이상이면 자동촬영

⚠️ **GitHub.com Fallback** ⚠️