I know it's a bit weird but yes.The idea was to make a real time image classifier ,so i went with this.
I downloaded around 100 images of samosa from the google images, though i knew one cannot make a deep learning model with that sort of data. hence to increase the dataset size i applied augmentation to the images via keras data generator function by augmenting its properties to a large extent
i included 5000 images of cifar -10 fr training the model on positive as well as negative examples and 2500 images for testing set.
i trained the model firstly by using the tensorflow and then keras(scikit of deep learning lol) . it gave pretty good results by the way
The real task was to make this real time . initially i thought to make use of laptop's webcam but it did'nt go well (Ps: the camera quality was stupid in fact) . Then i decided to use my phone's camera
So a friend of mine made a simple android app which clicks a photo and posts a downloadable link at the firebase and i used that link to straight away download the file and load the image in the model . i know it's not that much fancy but i had to do this