geraldzakwan/dl-comp-vis-final-project

dl-comp-vis-final-project

How to Reproduce the Result Using the Notebooks

Each notebook corresponds to one experiment mentioned in the report. For example, Landmark_Classification_ResNet_FineTune.ipynb is associated with training a ResNet18 using FineTune method, i.e. unfreezing all layers.

Before running the notebook, prepare the dataset using the following way:

1. Find the cell that contains below command:

!cp -r "drive/MyDrive/DL for Comp Vis/Final Project/data/landmark" data

2. If you run the notebook in Google Colab, then you can mount your Google Drive and change the path (after the -r option) into your desired data path in drive.

3. If you run it locally, change the path (after the -r option) into your desired local data path. Obviously, you can also place directly the dataset under the data directory and skip/comment the cp command.

4. Make sure that data directory has the same parent directory to that of the notebook and is structured like below.

data
  landmark
    train
      alma
      lion
      uris
    val
      alma
      lion
      uris
    test
      alma
      lion
      uris

Trying Interactive Demo using Tensorflow.js

You can try an interactive demo of this image classification app by visiting the below link.

https://dl-for-comp-vis-final-proj.herokuapp.com

Alternatively, if the above link doesn't work, you could run it locally by doing the following steps:

1. Get into the tfjs_model directory.

cd tfjs_model

2. Run the static web app using Python HTTP Server. Below example is using port 8282.

python3 -m http.server 8282

3. Visit the app by entering the URL: localhost:8282 in your browser, preferably Google Chrome, because that's where I tested the app.