documentation

step 1 :
    Data gathering .
    the file image_downloader.py has the code used to download images from bing website . 
    
step 2:
    data information and its labels .
    the file image_downloader has the code for in which it lists the contents of the folder 
    and inputs them in the csv file of each image folder .

step 3 : preparation of training set 
    So of the entire data . 70% of the data was used to train the model . the other 20% was used to validate 
    the next 10 % was used to test .
    After which data augmentation was on done on the training image where each photo accounted for 6 images in the final training data

Model . 
    there are 4 models available here .
    1.Build a CNN model from scratch-->cnn.py 
    2.Use pre-trained model (resnet-50) + finetune only the last layer-->pretrain_last_layer.py
    fc_inputs=resnet50.fc.in_features
    resnet50.fc = nn.Sequential(
#     nn.Linear(fc_inputs, 256),
#     nn.ReLU(),
#     nn.Dropout(0.4),
#     nn.Linear(256, 3), # Since 10 possible outputs
#     nn.LogSoftmax(dim=1) # For using NLLLoss()
     nn.Dropout(0.4),
     nn.Linear(fc_inputs, 3),
     nn.LogSoftmax(dim=1)
     
     used to retrain just the last layer 

    3.Use pre-trained model + finetune last few layers-->PreTrain_few.py
    model downloaded from torch vision
    resnet50 = models.resnet50(weights=models.ResNet50_Weights.DEFAULT)
    i=0
    for params in resnet50.parameters():
            params.requires_grad=False

    unfreeze_layers = ["layer4", "fc"]#these are last two layers of the resnet50 model 
    for name, param in resnet50.named_parameters():
        if any(layer_name in name for layer_name in unfreeze_layers):
            param.requires_grad = True
    in_features=resnet50.fc.in_features
    out_features=3
    resnet50.fc=nn.Linear(in_features,out_features) 
    

        the above represents the code used to the just allow train of the last 2 layers 

    4.Build ViT model--> Vit.py 
        model downloaded from torchvison 
    files :
    albumentations.py 
    used the albumentations library for data augmentation 

    main.py is just a file to be run if you you want to upload an image through a web server through an api . 
    uvicorn main:app --reload
    where you can upload the file and wait for the probabilities being displayed .
    Its a very simple demo of api integration . not a very sophisticated one .