import gradio as gr
from fastai.vision.all import *
import skimage
from facenet_pytorch import MTCNN
import torch
import pandas as pd
from PIL import Image, ImageDraw, ImageFont

device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
print('Running on device: {}'.format(device))

mtcnn = MTCNN(margin=40, keep_all=True, post_process=False, device=device)

learn = load_learner('export.pkl')

labels = learn.dls.vocab

emotions = {"x": [], "y": [], "State": []}

def predict(img):
    img = PILImage.create(img)
    boxes, _ = mtcnn.detect(img)
    o_img = img.copy()
    draw = ImageDraw.Draw(o_img)
    for box in boxes:
        coords = tuple(box.tolist())
        pred,pred_idx,probs = learn.predict(img.crop(coords))
        draw.rectangle(coords, outline=(0, 0, 0), width=1)
        draw.text((coords[0]-10, coords[1]-10), pred, font=ImageFont.truetype("arial"))
    return o_img
    '''
        emotions["x"].append((coords[0] + coords[2])/2)
        emotions["y"].append((coords[1] + coords[3])/2)
        emotions["State"].append(pred)
    emotions_df = pd.DataFrame(emotions)
    return gr.ScatterPlot.update(
        value=emotions_df,
        x="x",
        y="y",
        color="State",
        title="Class Heat Map",
        color_legend_title="State of the student",
        caption="Class Monitor",
    )
    '''
 
title = "Students emotion classifer"
description = "A students emotion classifer trained with fastai. Created as a demo for Gradio and HuggingFace Spaces."
interpretation='default'
enable_queue=True

#gr.Interface(fn=predict,inputs=gr.Image(source="webcam",shape=(512, 512)),outputs=gr.outputs.Label(num_top_classes=3),title=title,description=description,interpretation=interpretation,enable_queue=enable_queue).launch()
#gr.Interface(fn=predict,inputs=gr.Image(source="webcam",shape=(512, 512)),outputs=gr.ScatterPlot(),title=title,description=description,interpretation=interpretation,enable_queue=enable_queue, share=True).launch()
gr.Interface(fn=predict,inputs=gr.Image(),outputs=gr.Image(),title=title,description=description,interpretation=interpretation,enable_queue=enable_queue, share=True).launch()