import torch
from torchvision import transforms, models, datasets, utils

import gradio as gr

#from PIL import Image

model = models.densenet121()
model.load_state_dict(torch.load('derma_diseases_detection_best.pt', map_location=torch.device('cpu')))

from monai.transforms import (
    EnsureChannelFirst,
    Compose,
    LoadImage,
    ScaleIntensity,
)

test_transforms = Compose(
        [LoadImage(image_only=True), EnsureChannelFirst(), ScaleIntensity()]
    )

class_names = [
    'AbdomenCT', 'BreastMRI', 'CXR', 'ChestCT', 'Hand', 'HeadCT'
]

import os, glob

#examples_dir = './samples'
#example_files = glob.glob(os.path.join(examples_dir, '*.jpg'))

def classify_image(image_filepath):
    input = test_transforms(image_filepath)

    model.eval()
    with torch.no_grad():
        pred = model(input.unsqueeze(dim=0))
    
    prob = torch.nn.functional.softmax(pred[0], dim=0)
 
    confidences = {class_names[i]: float(prob[i]) for i in range(6)} 
    print(confidences)

    return confidences


with gr.Blocks(title="Medical Image Classification with MONAI - ClassCat",
            css=".gradio-container {background:mintcream;}"
        ) as demo:
    gr.HTML("""<div style="font-family:'Times New Roman', 'Serif'; font-size:16pt; font-weight:bold; text-align:center; color:royalblue;">Medical Image Classification with MONAI</div>""")

    with gr.Row(): 
        input_image = gr.Image(type="filepath", image_mode="L", shape=(64, 64))        
        output_label=gr.Label(label="Probabilities", num_top_classes=3)

    send_btn = gr.Button("Infer")
    send_btn.click(fn=classify_image, inputs=input_image, outputs=output_label)

demo.launch(debug=True)