import cv2
import torch
import numpy as np
import gradio as gr
from PIL import Image
from super_image import ImageLoader, EdsrModel, MsrnModel, MdsrModel, AwsrnModel, A2nModel, CarnModel, PanModel, \
    HanModel, DrlnModel, RcanModel

title = "super-image"
description = "State of the Art Image Super-Resolution Models."
article = "<p style='text-align: center'><a href='https://github.com/eugenesiow/super-image'>Github Repo</a>" \
          "| <a href='https://eugenesiow.github.io/super-image/'>Documentation</a> " \
          "| <a href='https://github.com/eugenesiow/super-image#scale-x2'>Models</a></p>"


def get_model(model_name, scale):
    if model_name == 'EDSR':
        model = EdsrModel.from_pretrained('eugenesiow/edsr', scale=scale)
    elif model_name == 'MSRN':
        model = MsrnModel.from_pretrained('eugenesiow/msrn', scale=scale)
    elif model_name == 'MDSR':
        model = MdsrModel.from_pretrained('eugenesiow/mdsr', scale=scale)
    elif model_name == 'AWSRN-BAM':
        model = AwsrnModel.from_pretrained('eugenesiow/awsrn-bam', scale=scale)
    elif model_name == 'A2N':
        model = A2nModel.from_pretrained('eugenesiow/a2n', scale=scale)
    elif model_name == 'CARN':
        model = CarnModel.from_pretrained('eugenesiow/carn', scale=scale)
    elif model_name == 'PAN':
        model = PanModel.from_pretrained('eugenesiow/pan', scale=scale)
    elif model_name == 'HAN':
        model = HanModel.from_pretrained('eugenesiow/han', scale=scale)
    elif model_name == 'DRLN':
        model = DrlnModel.from_pretrained('eugenesiow/drln', scale=scale)
    elif model_name == 'RCAN':
        model = RcanModel.from_pretrained('eugenesiow/rcan', scale=scale)
    else:
        model = EdsrModel.from_pretrained('eugenesiow/edsr-base', scale=scale)
    return model


def inference(img, scale_str, model_name):
    max_res = 1024
    scale = int(scale_str.replace('x', ''))
    width, height = img.size
    print(width, height)
    if width > max_res or height > max_res:
        img = img.thumbnail((max_res, max_res), Image.ANTIALIAS)
    model = get_model(model_name, scale)
    try:
        inputs = ImageLoader.load_image(img)
        preds = model(inputs)
        preds = preds.data.cpu().numpy()
        pred = preds[0].transpose((1, 2, 0)) * 255.0
        return Image.fromarray(pred.astype('uint8'), 'RGB')
    except Exception as e:
        print(e)
        return None


torch.hub.download_url_to_file('http://people.rennes.inria.fr/Aline.Roumy/results/images_SR_BMVC12/input_groundtruth/baby_mini_d3_gaussian.bmp',
                               'baby.bmp')
torch.hub.download_url_to_file('http://people.rennes.inria.fr/Aline.Roumy/results/images_SR_BMVC12/input_groundtruth/woman_mini_d3_gaussian.bmp',
                               'woman.bmp')
torch.hub.download_url_to_file('http://people.rennes.inria.fr/Aline.Roumy/results/images_SR_BMVC12/input_groundtruth/bird_mini_d4_gaussian.bmp',
                               'bird.bmp')

# models = ['EDSR-base', 'DRLN', 'EDSR', 'MDSR', 'A2N', 'PAN', 'AWSRN-BAM', 'MSRN']
models = ['EDSR-base', 'A2N', 'PAN', 'AWSRN-BAM', 'MSRN']
scales = [2, 3, 4]
for model_name in models:
    for scale in scales:
        get_model(model_name, scale)

gr.Interface(
    inference,
    [
        gr.inputs.Image(type="pil", label="Input"),
        gr.inputs.Radio(["x2", "x3", "x4"], label='scale'),
        gr.inputs.Dropdown(choices=models,
                           label='Model')
    ],
    gr.outputs.Image(type="pil", label="Output"),
    title=title,
    description=description,
    article=article,
    examples=[
        ['baby.bmp', 'x2', 'EDSR-base'],
        ['woman.bmp', 'x3', 'MSRN'],
        ['bird.bmp', 'x4', 'PAN']
    ],
    allow_flagging='never',
    ).launch(debug=False)