What it does
It takes a scrambled image, and returns a string representing the ordering of the pieces that is required to unscramble this image.
How we built it
First, divide each image into 4 equal rectangular pieces. Then pass each piece through the CNN. CNN extracts useful features and outputs a feature vector. We concatenate all 4 feature vectors into one using the Flatten layer. Then we pass this combined vector through a feed-forward network. The last layer of this network gives us a 16-unit long vector. We reshape this 16-unit vector into a matrix of 4x4.
Challenges we ran into
The dataset was huge with about 50k images, hence we had to use a tensor slicer to tackle this problem.
What's next for Puzzle Solver
Make it dynamic so it works for all input sizes.
Built With
- keras
- python
Log in or sign up for Devpost to join the conversation.