app.py

import gradio as gr
from transformers import SegformerFeatureExtractor, SegformerForSemanticSegmentation
import matplotlib.pyplot as plt
from matplotlib import gridspec
import numpy as np
from PIL import Image
import tensorflow as tf
import requests

# Load the pre-trained model and feature extractor
feature_extractor = SegformerFeatureExtractor.from_pretrained("nvidia/segformer-b0-finetuned-cityscapes-640-1280")
model = SegformerForSemanticSegmentation.from_pretrained("nvidia/segformer-b0-finetuned-cityscapes-640-1280")

def my_palette():
    return [
        [131, 162, 255],
        [180, 189, 255],
        [255, 227, 187],
        [255, 210, 143],
        [248, 117, 170],
        [255, 223, 223],
        [255, 246, 246],
        [174, 222, 252],
        [150, 194, 145],
        [255, 219, 170],
        [244, 238, 238],
        [50, 38, 83],
        [128, 98, 214],
        [146, 136, 248],
        [255, 210, 215],
        [255, 152, 152],
        [162, 103, 138],
        [63, 29, 56],
        [0,0,0]
    ]

labels_list = []

with open(r"labels.txt", "r") as fp:
    for line in fp:
        labels_list.append(line[:-1])

colormap = np.asarray(my_palette())

def greet(input_img):
    inputs = feature_extractor(images=input_img, return_tensors="pt")
    outputs = model(**inputs)
    logits = outputs.logits

    logits_tf = tf.transpose(logits.detach(), [0, 2, 3, 1])

    logits_tf = tf.image.resize(
        logits_tf, [640, 1280]
    )
    seg = tf.math.argmax(logits_tf, axis=-1)[0]

    color_seg = label_to_color_image(seg.numpy())

    # Resize color_seg to match the shape of input_img
    color_seg_resized = tf.image.resize(color_seg, (input_img.shape[0], input_img.shape[1]))

    pred_img = np.array(input_img) * 0.5 + color_seg_resized * 0.5

    # Convert pred_img to NumPy array and then change data type
    pred_img = np.array(pred_img).astype(np.uint8)

    fig = draw_plot(pred_img, seg.numpy())
    return fig

def draw_plot(pred_img, seg):
    fig = plt.figure(figsize=(20, 15))
    grid_spec = gridspec.GridSpec(1, 2, width_ratios=[6, 1])

    plt.subplot(grid_spec[0])
    plt.imshow(pred_img)
    plt.axis("off")

    LABEL_NAMES = np.asarray(labels_list)
    FULL_LABEL_MAP = np.arange(len(LABEL_NAMES)).reshape(len(LABEL_NAMES), 1)
    FULL_COLOR_MAP = label_to_color_image(FULL_LABEL_MAP)

    # Limit unique_labels to be within the range of colormap
    unique_labels = np.unique(seg.astype("uint8"))
    unique_labels = unique_labels[unique_labels < len(FULL_COLOR_MAP)]

    ax = plt.subplot(grid_spec[1])

    if len(unique_labels) > 0:
        plt.imshow(FULL_COLOR_MAP[unique_labels].astype(np.uint8), interpolation="nearest")
        ax.yaxis.tick_right()
        plt.yticks(range(len(unique_labels)), LABEL_NAMES[unique_labels])
    else:
        # Handle case when there are no unique labels
        plt.imshow(np.zeros((1, 1, 3), dtype=np.uint8))
        ax.yaxis.tick_right()
        plt.yticks([], [])

    plt.xticks([], [])
    ax.tick_params(width=0.0, labelsize=25)
    return fig
def label_to_color_image(label):
    if label.ndim != 2:
        raise ValueError("Expect 2-D input label")

    # Clip label values to be within the range of colormap
    label = np.clip(label, 0, len(colormap) - 1)
    return colormap[label]

iface = gr.Interface(
    fn=greet,
    inputs="image",
    outputs=["plot"],
    examples=["image (1).jpg", "image (2).jpg", "image (3).jpg", "image (4).jpg", "image (5).jpg"],
    allow_flagging="never"
)
iface.launch(share=True)