app.py

import numpy as np
import cv2
import gradio as gr
import torch

from ade20k_colors import colors

from transformers import BeitFeatureExtractor, BeitForSemanticSegmentation


beit_models = ['microsoft/beit-base-finetuned-ade-640-640',
               'microsoft/beit-large-finetuned-ade-640-640']

models = [BeitForSemanticSegmentation.from_pretrained(m) for m in beit_models]
extractors = [BeitFeatureExtractor.from_pretrained(m) for m in beit_models]


def apply_colors(img):
    ret = np.zeros((img.shape[0], img.shape[1], 3), dtype=np.uint8)
 
    for y in range(img.shape[0]):
        for x in range(img.shape[1]):
            ret[y,x] = colors[np.argmax(img[y,x])]

    return ret


def inference(image, chosen_model):
    feature_extractor = extractors[chosen_model]
    model = models[chosen_model]

    inputs = feature_extractor(images=image, return_tensors='pt')
    outputs = model(**inputs)

    logits = outputs.logits

    output = torch.sigmoid(logits).detach().numpy()[0]
    output = np.transpose(output, (1,2,0))

    output = apply_colors(output)

    return cv2.resize(output, image.shape[1::-1])


inputs = [gr.inputs.Image(label='Input Image'),
          gr.inputs.Radio(['Base', 'Large'], label='BEiT Model', type='index')]

gr.Interface(
    inference, 
    inputs,
    gr.outputs.Image(label='Output'),
    title='BEiT - Semantic Segmentation',
    description='BEIT: BERT Pre-Training of Image Transformers',
    examples=[['images/armchair.jpg', 'Base'],
              ['images/cat.jpg', 'Base'],
              ['images/plant.jpg', 'Large']]
    ).launch()