Google Landmark Recognition Challenge
- We use TensorFlow for image recognition. We will be using transfer learning, which means we are starting with a model that has been already trained on another problem. We will then be retraining it on a similar problem. Deep learning from scratch can take days, but transfer learning can be done in short order.
- MoBileNet will be used, to be small and efficient.
- Follow tutorial from codelabs
- Clone the respo from tensorflow
- After done the tutorial, replace the label_image.py from this respo to the cloned respo
- copy the run.ipynb to cloned respo and run
- image_downloader image downloader for both testing and training data, with compression option
- datasets analysis data analysis and visualisation
- label_image.py for image labelling and prediction
- run.ipynb to run the prediction
- train: 336 GB with 1,220,165 images
- test: 34.9 GB with 116,163 images
- May want to consider run on Cloud