Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,74 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from transformers import DPTFeatureExtractor, DPTForDepthEstimation
|
3 |
+
import torch
|
4 |
+
import numpy as np
|
5 |
+
from PIL import Image
|
6 |
+
import cv2
|
7 |
+
from sklearn.cluster import KMeans
|
8 |
+
from matplotlib import pyplot as plt
|
9 |
+
|
10 |
+
torch.hub.download_url_to_file('http://images.cocodataset.org/val2017/000000039769.jpg', 'cats.jpg')
|
11 |
+
|
12 |
+
feature_extractor = DPTFeatureExtractor.from_pretrained("Intel/dpt-large")
|
13 |
+
model = DPTForDepthEstimation.from_pretrained("Intel/dpt-large")
|
14 |
+
|
15 |
+
def process_image(image):
|
16 |
+
# Prepare image for the model
|
17 |
+
encoding = feature_extractor(image, return_tensors="pt")
|
18 |
+
|
19 |
+
# Forward pass
|
20 |
+
with torch.no_grad():
|
21 |
+
outputs = model(**encoding)
|
22 |
+
predicted_depth = outputs.predicted_depth
|
23 |
+
|
24 |
+
# Interpolate to original size
|
25 |
+
prediction = torch.nn.functional.interpolate(
|
26 |
+
predicted_depth.unsqueeze(1),
|
27 |
+
size=image.size[::-1],
|
28 |
+
mode="bicubic",
|
29 |
+
align_corners=False,
|
30 |
+
).squeeze()
|
31 |
+
depth_map_gray = (prediction.cpu().numpy() * 255).astype('uint8')
|
32 |
+
|
33 |
+
# Perform feature segmentation
|
34 |
+
rgb_image = np.array(image)
|
35 |
+
depth_threshold = 1000
|
36 |
+
binary_mask = np.where(depth_map_gray > depth_threshold, 255, 0).astype(np.uint8)
|
37 |
+
gray_image = cv2.cvtColor(rgb_image, cv2.COLOR_RGB2GRAY)
|
38 |
+
pixels = gray_image.reshape((-1, 1))
|
39 |
+
num_clusters = 3
|
40 |
+
kmeans = KMeans(n_clusters=num_clusters)
|
41 |
+
kmeans.fit(pixels)
|
42 |
+
labels = kmeans.labels_
|
43 |
+
labels = labels.reshape(gray_image.shape)
|
44 |
+
cluster_features = []
|
45 |
+
for i in range(num_clusters):
|
46 |
+
mask = np.where(labels == i, 255, 0).astype(np.uint8)
|
47 |
+
cluster_image = cv2.bitwise_and(rgb_image, rgb_image, mask=mask)
|
48 |
+
cluster_features.append(cluster_image)
|
49 |
+
|
50 |
+
# Prepare output images
|
51 |
+
depth_image = Image.fromarray(depth_map_gray, mode='L')
|
52 |
+
cluster_images = [Image.fromarray(cluster) for cluster in cluster_features]
|
53 |
+
|
54 |
+
return depth_image, cluster_images
|
55 |
+
|
56 |
+
title = "Demo: zero-shot depth estimation with DPT and feature segmentation"
|
57 |
+
description = "Demo for Intel's DPT with feature segmentation, a Dense Prediction Transformer for state-of-the-art dense prediction tasks such as semantic segmentation and depth estimation."
|
58 |
+
examples = [['cats.jpg']]
|
59 |
+
|
60 |
+
iface = gr.Interface(
|
61 |
+
fn=process_image,
|
62 |
+
inputs=gr.inputs.Image(type="pil"),
|
63 |
+
outputs=[
|
64 |
+
gr.outputs.Image(type="pil", label="predicted depth"),
|
65 |
+
gr.outputs.Image(type="pil", label="cluster 1"),
|
66 |
+
gr.outputs.Image(type="pil", label="cluster 2"),
|
67 |
+
gr.outputs.Image(type="pil", label="cluster 3"),
|
68 |
+
],
|
69 |
+
title=title,
|
70 |
+
description=description,
|
71 |
+
examples=examples,
|
72 |
+
enable_queue=True
|
73 |
+
)
|
74 |
+
iface.launch(debug=True)
|