- A mobile android application that uses Deeplearning to recognize images in real-time taken by the mobile phone's camera.
- This project is maintained by 정명지, 오서영, 강성원
- Our Team name is "Trinity"
- Jul. 6, 2020 ~ Sep. 2, 2020
Developer | Individual Role | - |
---|---|---|
정명지 | Basic bio study, image crawling | |
오서영 | Basic Machine learning presentation - Overview | Presentation |
강성원 | Basic JAVA presentation - Overview |
Developer | Individual Role | - |
---|---|---|
정명지 | Handwriting application structure design | Rough design |
오서영 | Basic ML presentation - Image classification with CNN | Presentation |
강성원 | Basic JAVA presentation - Android Studio |
Developer | Individual Role | - |
---|---|---|
정명지 | Data collection - flower dataset (image) | |
오서영 | Mobile app design - icon, color, view | Main screen design |
강성원 | Mobile app - touch slide motion event |
Developer | Individual Role | - |
---|---|---|
정명지 | Data collection - flower dataset (common/scientific name) | In-app dataset |
오서영 | Banner design, Baseline CNN with flower image dataset | Banner design |
강성원 | Mobile app - UI relocation, touch event |
Developer | Individual Role | - |
---|---|---|
정명지 | Google Image crawling for training | |
오서영 | Single layer NN Presentation for study, Resnet | Presentation |
강성원 | Android Studio Presentation for study, Mobile app - Animation, in-app data insertion |
Developer | Individual Role | - |
---|---|---|
정명지 | Google Image crawling for training | |
오서영 | Model Selection to complement accuracy, Page Design | Design |
강성원 | Mobile app - Add new pages with page design |
Developer | Individual Role | - |
---|---|---|
정명지 | Study AI and JAVA for report | |
오서영 | Test with a sample | Test |
강성원 | Mobile app - Implement camera, build data models |
Developer | Individual Role | - |
---|---|---|
정명지 | Write up report | |
오서영 | Connecting Tensorflow Models to Android Studio, Icon Design | Icon |
강성원 | Connecting Tensorflow Models to Android Studio |
Applying Tensorflow to Android
- Convert h5 to pb
- Convert pb to tflite
- Make test.py file | Code
- Tensorflow Flower Dataset : 5 classes (daisy, dandelion, roses, sunflowers, tulips)
[1] tf_flowers, https://www.tensorflow.org/datasets/catalog/tf_flowers
1. Baseline CNN (100 iterations, 32 batch) | Code
Train accuracy: 85.62%
Val accuracy: 69.38%
2. Resnet (50 iterations, 32 batch) | Code
Train accuracy : 85.00%
Val accuracy : 66.25%
4685 training set with 5 class | Code
Train accuracy: 72.19%
Val accuracy: 70.00%
Train accuracy: 83.13%
Val accuracy: 73.12%
Train accuracy: 95.00%
Val accuracy: 73.75%
Train accuracy: 79.37%
Val accuracy: 65.62%
Train accuracy: 80.00%
Val accuracy: 80.00%
[1] Advanced Computer Vision with TensorFlow, https://stephan-osterburg.gitbook.io/coding/coding/ml-dl/tensorfow