Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import matplotlib.pyplot as plt
|
3 |
+
import torch
|
4 |
+
import numpy as np
|
5 |
+
from transformers import DPTImageProcessor, DPTForDepthEstimation
|
6 |
+
from PIL import Image
|
7 |
+
import requests
|
8 |
+
|
9 |
+
# Load model and processor
|
10 |
+
st.title("Depth Estimation using DPT")
|
11 |
+
st.write("Upload an image to estimate its depth map.")
|
12 |
+
|
13 |
+
@st.cache_resource
|
14 |
+
def load_model():
|
15 |
+
processor = DPTImageProcessor.from_pretrained("Intel/dpt-large")
|
16 |
+
model = DPTForDepthEstimation.from_pretrained("Intel/dpt-large")
|
17 |
+
return processor, model
|
18 |
+
|
19 |
+
processor, model = load_model()
|
20 |
+
|
21 |
+
# File uploader
|
22 |
+
uploaded_file = st.file_uploader("Upload an image", type=["jpg", "png", "jpeg"])
|
23 |
+
|
24 |
+
if uploaded_file is not None:
|
25 |
+
image = Image.open(uploaded_file)
|
26 |
+
st.image(image, caption="Uploaded Image", use_column_width=True)
|
27 |
+
|
28 |
+
# Process image
|
29 |
+
inputs = processor(images=image, return_tensors="pt")
|
30 |
+
with torch.no_grad():
|
31 |
+
outputs = model(**inputs)
|
32 |
+
predicted_depth = outputs.predicted_depth
|
33 |
+
|
34 |
+
# Interpolate to original size
|
35 |
+
prediction = torch.nn.functional.interpolate(
|
36 |
+
predicted_depth.unsqueeze(1),
|
37 |
+
size=image.size[::-1],
|
38 |
+
mode="bicubic",
|
39 |
+
align_corners=False,
|
40 |
+
)
|
41 |
+
|
42 |
+
# Convert to NumPy array
|
43 |
+
output = prediction.squeeze().cpu().numpy()
|
44 |
+
normalized_depth = (output - output.min()) / (output.max() - output.min()) # Normalize to [0, 1]
|
45 |
+
|
46 |
+
# Plot the results
|
47 |
+
fig, ax = plt.subplots(1, 2, figsize=(12, 6))
|
48 |
+
ax[0].imshow(image)
|
49 |
+
ax[0].set_title("Original Image")
|
50 |
+
ax[0].axis("off")
|
51 |
+
ax[1].imshow(normalized_depth, cmap="inferno")
|
52 |
+
ax[1].set_title("Predicted Depth Map")
|
53 |
+
ax[1].axis("off")
|
54 |
+
|
55 |
+
# Display result
|
56 |
+
st.pyplot(fig)
|