Inspiration

rag tag team who's visionary ideas hope to change the world

What it does

Predicts the focal length of the image that was used to take the picture. Has a crowdsourcing component where users can help train the model by uploading an image and given focal length (label).

How we built it

We used the Tensorflow Inception v3 neural network trained on the ImageNet data set and modified the top of the network to classify to new data we trained it on using Transfer Learning.

Challenges we ran into

Tensorflow :(

Accomplishments that we're proud of

Trained a model using Transfer Learning on the Inception v3 convolutional neural network that's over 60% accurate in predictions trained with a small dataset.

What we learned

How Transfer Learning methods work to more easily train a smaller data set to create new classifications without retraining the entire model on a large data set (e.g. ImageNet).

What's next for PerspectiveML

Continue gathering more training data and train a better model that generalizes well to new data. We want to use the focal length and known field of view to compute the approximate distance from the subject and the camera.

Built With

Share this project:

Updates