Vriational Autoencoder Application: Predction of Trajectory

What it does

The model we built can successfully predict the trajectory of silver particles in the first 100 frames with very little uncertainties and sufficient accuracy.

How we built it

The process of building this model was not so easy as it seems. After importing and cleaning the data, we tried several combinations of AutoEncoders and layers.

Challenges we ran into

The CPU on our device is not powerful enough to import all the data with dataloader at the beginning. However, we find out our way to import all the wanted data by creating a class DataSet

Accomplishments that we're proud of

We are proud of running the code on our normal laptops with such a gigantic amount of data, as well as the fact that we successfully trained the model with such good performance.

What we learned

In 2022 AiHack, the experience we had is going to be such a treasure in our lives. It is the bridge and base to build the friendship between our members. More importantly, this is a chance to learn the power of a team can be way more than an individual. Therefore, it is so important to learn how to work in a team with maximum productivity.

What's next for Variational Autoencoder Application: Prediction of Trajectory

If we have more time to improve our project, we would like to work on a better device so that we can compare the performance between all valid combinations. It can also help us to work out a more efficient model based on this prototype.

Built With

• jupyternotebook
• python
• vae