Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import requests
|
3 |
+
import torch
|
4 |
+
from PIL import Image
|
5 |
+
from transformers import AutoImageProcessor, Mask2FormerForUniversalSegmentation
|
6 |
+
import numpy as np
|
7 |
+
import matplotlib.pyplot as plt
|
8 |
+
|
9 |
+
# Load Mask2Former fine-tuned on ADE20k semantic segmentation
|
10 |
+
st.title("Mask2Former Semantic Segmentation")
|
11 |
+
st.write("Upload an image to perform semantic segmentation using Mask2Former.")
|
12 |
+
|
13 |
+
processor = AutoImageProcessor.from_pretrained("facebook/mask2former-swin-large-ade-semantic")
|
14 |
+
model = Mask2FormerForUniversalSegmentation.from_pretrained("facebook/mask2former-swin-large-ade-semantic")
|
15 |
+
|
16 |
+
def segment_image(image: Image.Image):
|
17 |
+
inputs = processor(images=image, return_tensors="pt")
|
18 |
+
with torch.no_grad():
|
19 |
+
outputs = model(**inputs)
|
20 |
+
predicted_semantic_map = processor.post_process_semantic_segmentation(outputs, target_sizes=[image.size[::-1]])[0]
|
21 |
+
return predicted_semantic_map
|
22 |
+
|
23 |
+
def visualize_segmentation(image: Image.Image, segmentation_map):
|
24 |
+
plt.figure(figsize=(10, 5))
|
25 |
+
plt.subplot(1, 2, 1)
|
26 |
+
plt.imshow(image)
|
27 |
+
plt.axis("off")
|
28 |
+
plt.title("Original Image")
|
29 |
+
|
30 |
+
plt.subplot(1, 2, 2)
|
31 |
+
plt.imshow(segmentation_map, cmap="jet", alpha=0.7)
|
32 |
+
plt.axis("off")
|
33 |
+
plt.title("Segmented Image")
|
34 |
+
|
35 |
+
st.pyplot(plt)
|
36 |
+
|
37 |
+
# File uploader for user to upload an image
|
38 |
+
uploaded_file = st.file_uploader("Upload an image", type=["jpg", "png", "jpeg"])
|
39 |
+
if uploaded_file:
|
40 |
+
image = Image.open(uploaded_file).convert("RGB")
|
41 |
+
st.image(image, caption="Uploaded Image", use_column_width=True)
|
42 |
+
|
43 |
+
if st.button("Segment Image"):
|
44 |
+
st.write("Processing the image...")
|
45 |
+
segmentation_map = segment_image(image)
|
46 |
+
visualize_segmentation(image, segmentation_map.numpy())
|
47 |
+
|
48 |
+
# Option to test with a sample image
|
49 |
+
if st.button("Use Sample Image"):
|
50 |
+
url = "http://images.cocodataset.org/val2017/000000039769.jpg"
|
51 |
+
image = Image.open(requests.get(url, stream=True).raw)
|
52 |
+
st.image(image, caption="Sample Image", use_column_width=True)
|
53 |
+
|
54 |
+
st.write("Processing the image...")
|
55 |
+
segmentation_map = segment_image(image)
|
56 |
+
visualize_segmentation(image, segmentation_map.numpy())
|