Sparsity-Invariant-CNNs

This repository contains all the files for you to implement Sparsity Invariant CNN. In order to run this project you need to make a Google Drive folder to upload these files in order to work with Google Colab. Open CNN_v_s_Sparse_CNN_Tf.ipynb to run the program.

How to Setup?

1. Make a folder train in our project ROOT directory.
2. Prepare the dataset now by downloading the velodyne raw data extracting it in the train folder.
3. After dowloading the raw_input data download the groundtruth and extract it in the same train folder.
4. Make a folder NNFL Assignment in your ROOT_DIR of Google Drive and upload all the files from your project ROOT directory.

Using a local runtime

Change the ROOT_DIR to the location of your project root in order to run the notebook in a local runtime. We have strictly used Google Colab and Google Drive for developing, running and troubleshooting the whole project so we can't guarantee on using a local runtime.

Additional Information

1. The checkpoints and log files for different models training will be stored in Models folder and in-depth statistics can be found in the Statistics folder.
2. Already trained model names can be found in models.txt file in Models folder however the variables have been initialized in the notebook of already trained models.
3. The graphs and outputs of the network are stored in Graphs and Output folder. Graphs and Output\multi_input_and_output_model.png contain the flowchart of the network and can be referred to get an idea of the network.