Inspiration
Application of the AI algorithms to the world of fashion. Improving the data selection process and quality of the result by mixing more different types of AI algorithms.
What it does
- Preprocessing: improving data quality. Deletion of the outliers. Imputing variables.
- Related elements search creation of related elements communities. Input labeling for the tabular data.
- Image features extraction: Image classification. Input labeling for the graphical data.
- NN training
- Testing
How we built it
Multimodal embedding. Triplet metric learning for tabular data. VGG16 convolutional network for the image features extraction.
Challenges we ran into
Selection of the positive sample for the application of triplet metric learning for tabular data classification. We solved with the definition of a dependencies graph and the communities analysis. Technical implementation of the training of the NN.
Accomplishments that we're proud of
Definition of a trained model as the resulting relationship between two different AI models. Doing a premise research of the method to fit the data. Trying different architectures.
What we learned
Practicals with AI algorithms, teamwork, and applying NN to real-life problems.
What's next for MANGO - AIpe - Multimodal embedding fashion compatibility
Better training (more epochs) and an improved visualization.
Documentation
The documentation is available in the .ipynb files in the code folders of the repository.
Authors
Gianluca Graziadei, Pau Amargant, David Gallardo Quesada, Carlos Arbonés Sotomayor
Log in or sign up for Devpost to join the conversation.