This project predicts logos of brands using doodles as input. It uses a Webcam to capture video and tracks path of specific coloured objects to draw the doodle on the webcam output window.
Logos used:
- Adidas
- Apple
- BMW
- Citroen
- FedEx
- HP
- Mcdonald's
- Nike
- Pepsi
- Puma
MobileNet CNN architecture was used. The last 18 layers were retrained on the dataset to benefit from transfer learning
Flickr-27 dataset was used. To generate more data, the images from flickr-27 were converted to binary images and data augmentation was also applied.
- data_conv.py: python programme to convert RGB training images to binary images for better feature extraction.
- data_structuring.py: python programme to structure data processed by data_conv.py to folder respective of the brands so as to facilitate data augmentation and training.(uses text files extracted from flickr27 datset)
- mob_logo_model.h5: trained model
- screen_reader.py: Main application programme. Run this programme to launch application
- train_logo.ipynb: Notebook for training of model.
.gif){kind=logo}
.gif){kind=logo}