🔥🔥 Train DGCNN Model using PyTorch Geometric and Weights & Biases 🪄🐝

This notebook demonstrates an implementation of the Dynamic Graph CNN for point cloud segmnetation implemented using PyTorch Geometric and experiment tracked and visualized using Weights & Biases. The code here is inspired by this original implementation.

If you wish to know how to evaluate the model on the ShapeNetCore dataset using Weights & Biases, you can check out the following notebook:

Install Required Packages

Import Libraries

Initialize Weights & Biases

We need to call wandb.init() once at the beginning of our program to initialize a new job. This creates a new run in W&B and launches a background process to sync data.

Load ShapeNet Dataset using PyTorch Geometric

We now load, preprocess and batch the ModelNet dataset for training, validation/testing and visualization.

Now, we need to offset the segmentation labels

Implementing the DGCNN Model using PyTorch Geometric

Training DGCNN and Logging Metrics on Weights & Biases

Next, you can check out the following notebook to learn how to evaluate the model on the ShapeNetCore dataset using Weights & Biases, you can check out the following notebook: