CS 5542 BigData Lab Report #10
Retrain Inception Model Final Layer for Image Dataset not covered in class. Report accuracy + Visualizations(Tensor Board) + Report Confusion Matrix for training and validation/testing.
- Use pre-trained Inception Model v3 to retrain different filter size of kernel in the pooling layer.
- Remove the final layer and train a new one based on our own datasets ( Kansas City Fountains ).
- Calculate bottleneck values for each image. --> To perform classification before the final output layer.
- Training ( % of images used in current training ) + Validation Accuracy ( % of correct label images ) + Cross-entropy ( how well the learning process is ) are calculated after each 10 iterations.
- The accuracy greatly enhanced compared to using only CNN model!
- Accuracy
- Retrain Accuracy
- Activation
- Cross-Entropy
- Final Training Ops
- I used the
gunicornto activate label_image.py to connect with Tensorflow API. When clicking the predict button through the web page, predictions of the image content is run in python tensorflow. Images are the 3 fountains in Kansas City that are used in the group project. The web is giving me predictions of my fountains but somehow the images aren't displaying!.. I would need to clear what's going on with that.
