from train import AnimeSegmentation
import cv2
import numpy as np
from loadimg import load_img
import gradio as gr
# import spaces
import torch

model = AnimeSegmentation.from_pretrained("skytnt/anime-seg") 

device = "cuda" if torch.cuda.is_available() else "cpu"
model.eval()
model.to(device)
img_size = model._hub_mixin_config["img_size"]


def get_mask(model, input_img, use_amp=True, s=640):
    input_img = (input_img / 255).astype(np.float32)
    h, w = h0, w0 = input_img.shape[:-1]
    h, w = (s, int(s * w / h)) if h > w else (int(s * h / w), s)
    ph, pw = s - h, s - w
    img_input = np.zeros([s, s, 3], dtype=np.float32)
    img_input[ph // 2:ph // 2 + h, pw // 2:pw // 2 + w] = cv2.resize(input_img, (w, h))
    img_input = np.transpose(img_input, (2, 0, 1))
    img_input = img_input[np.newaxis, :]
    tmpImg = torch.from_numpy(img_input).type(torch.FloatTensor).to(model.device)
    with torch.no_grad():
        if use_amp:
            with amp.autocast():
                pred = model(tmpImg)
            pred = pred.to(dtype=torch.float32)
        else:
            pred = model(tmpImg)
        pred = pred.cpu().numpy()[0]
        pred = np.transpose(pred, (1, 2, 0))
        pred = pred[ph // 2:ph // 2 + h, pw // 2:pw // 2 + w]
        pred = cv2.resize(pred, (w0, h0))[:, :, np.newaxis]
        return pred

# @spaces.GPU
def process(img):
    path = load_img(img,output_type="str")
    img = cv2.cvtColor(cv2.imread(path, cv2.IMREAD_COLOR), cv2.COLOR_BGR2RGB)
    mask = get_mask(model, img, use_amp= False, s=img_size)
    img = np.concatenate((img, mask * img, mask.repeat(3, 2) * 255), axis=1).astype(np.uint8)
    out = load_img(img)
    return out 


    
    
demo = gr.Interface(process,"image","image")

demo.launch(debug=True)