Inspiration
What it does
Predict the probability of a surface be damaged in future
How we built it
Challenges we ran into
Incomplete data, noisy data
Accomplishments that we're proud of
Firstly, we generated data samples from original ones to simulate the fracture growth during a timeline. Next, we trained a probabilistic model to quantify the certainty of fracture prediction. Since, this model is trained on the temporal data, the fracture labeling results are more reliable compared to the models working on single images.
What we learned
Handling incomplete data
What's next for Fracture Predictor
Finetuning the model on a larger dataset. Including the healthy data samples (without fracture) instead of generating and augmenting them.
