How do you train a machine learning model with only one image?
Our project successfully implements a vision recognition system using Augmentor and a Convolutional Neural Network
- Frontend: HTML, CSS, Bootstrap
- Backend: Flask, Python, Open CV, Google Cloud Platform
- Uses object detection and bounding box code for mural image detection with the power of OpenCV.
- Implements Augmentor to help supplement our sparse data set.
- Utilized transfer learning from TensorFlow Art as an artwork classifier.
- Wrote custom network for mural categorization. We built our model with NumPy and Tensorflow. In particular, we implemented a dual encoder convolutional neural network for unsupervised feature comparision. This essentially builds a "mural vector space" allowing the model, allowing for any murals to be quickly lookup and compared.
- Used Tensorboard for training validation and great visualizations.