#全景分割
from PIL import Image
import io
import matplotlib.pyplot as plt
import torch
import torchvision.transforms as T
import numpy
import gradio as gr
import itertools
import seaborn as sns
from panopticapi.utils import rgb2id


# These are the COCO classes
CLASSES = [
    'N/A', 'person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus',
    'train', 'truck', 'boat', 'traffic light', 'fire hydrant', 'N/A',
    'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse',
    'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe', 'N/A', 'backpack',
    'umbrella', 'N/A', 'N/A', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis',
    'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove',
    'skateboard', 'surfboard', 'tennis racket', 'bottle', 'N/A', 'wine glass',
    'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple', 'sandwich',
    'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake',
    'chair', 'couch', 'potted plant', 'bed', 'N/A', 'dining table', 'N/A',
    'N/A', 'toilet', 'N/A', 'tv', 'laptop', 'mouse', 'remote', 'keyboard',
    'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'N/A',
    'book', 'clock', 'vase', 'scissors', 'teddy bear', 'hair drier',
    'toothbrush'
]

# Detectron2 uses a different numbering scheme, we build a conversion table
coco2d2 = {}
count = 0
for i, c in enumerate(CLASSES):
  if c != "N/A":
    coco2d2[i] = count
    count+=1


# Draw the bounding boxes on image.
def fig2img(fig):
    buf = io.BytesIO()
    fig.savefig(buf)
    buf.seek(0)
    img = Image.open(buf)
    return img


# Draw the bounding boxes.
def visualize_prediction(result):
    palette = itertools.cycle(sns.color_palette())

    # The segmentation is stored in a special-format png
    panoptic_seg = Image.open(io.BytesIO(result['png_string']))
    panoptic_seg = numpy.array(panoptic_seg, dtype=numpy.uint8).copy()
    # We retrieve the ids corresponding to each mask
    panoptic_seg_id = rgb2id(panoptic_seg)

    # Finally we color each mask individually
    panoptic_seg[:, :, :] = 0
    for id in range(panoptic_seg_id.max() + 1):
        panoptic_seg[panoptic_seg_id == id] = numpy.asarray(next(palette)) * 255
    plt.figure(figsize=(15,15))
    plt.imshow(panoptic_seg)
    plt.axis('off')
    return fig2img(plt.gcf())

model, postprocessor = torch.hub.load('facebookresearch/detr', 
    'detr_resnet101_panoptic', 
    pretrained=True, 
    return_postprocessor=True, 
    num_classes=250)

transform = T.Compose([
    T.Resize(800),
    T.ToTensor(),
    T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
model.eval();


def detect_objects(image_input):
    model.eval();

    # if image comes from upload
    if image_input:
        image = image_input

    # mean-std normalize the input image (batch-size: 1)
    img = transform(image).unsqueeze(0)
    out = model(img)

    # the post-processor expects as input the target size of the predictions (which we set here to the image size)
    result = postprocessor(out, torch.as_tensor(img.shape[-2:]).unsqueeze(0))[0]
    
    #Visualize prediction
    viz_img = visualize_prediction(result)
    
    return viz_img
        
    
demo = gr.Interface(detect_objects, 
                    inputs = gr.Image(type='pil'), 
                    outputs = gr.Image(shape=(650,650)),
                    title = "全景分割",
                    allow_flagging="never",
                    examples = ['1.jpg', '2.jpg'])


if __name__ == '__main__':
    demo.queue().launch(server_name = "0.0.0.0", server_port = 7003, max_threads=40)