All the code is available in the CV.ipynb file
The pretrained model is by the name "model" ("note there is no extension to the name")

The initial a look at the dataset showed that the images are available in a folder and files having the names as links
The data set was loded in memory using numpy and pandas and resized to size (64,64,3)
The data generator uses data augmentation which helps increase the size of the dataset.
The train set also includes labels of test images which was filtered out


These two dictionaries are used to convert from labels to id and vice versa

label_to_id = {'Food':0, 'Attire':1, 'Decorationandsignage':2, 'misc':3}

id_to_label = {0: 'Food', 1:'Attire',2: 'Decorationandsignage',3: 'misc'}

The final four models are showed in the notebook

The model 2 was used to generate the submission.csv