Spaces:
Runtime error
Runtime error
Commit
•
5a4b7bc
1
Parent(s):
edb8a3c
Functional Sample
Browse files- app.py +53 -25
- examples/00015_colors.png +0 -0
- examples/car.JPG +0 -0
- examples/dash.jpg +0 -0
- utils.py +2 -2
app.py
CHANGED
@@ -6,6 +6,8 @@ import numpy as np
|
|
6 |
|
7 |
from huggingface_hub import from_pretrained_keras
|
8 |
|
|
|
|
|
9 |
custom_objects = {'BilinearUpSampling2D': BilinearUpSampling2D, 'depth_loss_function': None}
|
10 |
print('Loading model...')
|
11 |
model = from_pretrained_keras("keras-io/monocular-depth-estimation", custom_objects=custom_objects, compile=False)
|
@@ -15,56 +17,82 @@ import importlib
|
|
15 |
import utils
|
16 |
importlib.reload(utils)
|
17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
def infer(image, min_th, max_th):
|
19 |
-
print('_'*20)
|
20 |
inputs = utils.load_images([image])
|
21 |
outputs = utils.predict(model, inputs)
|
22 |
|
23 |
plasma = plt.get_cmap('plasma')
|
24 |
rescaled = outputs[0][:, :, 0]
|
25 |
-
print("Min Max Bef", np.min(rescaled), np.max(rescaled))
|
26 |
rescaled = rescaled - np.min(rescaled)
|
27 |
rescaled = rescaled / np.max(rescaled)
|
28 |
|
29 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
|
31 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
|
33 |
-
print("Shape Scaled:",rescaled.shape)
|
34 |
-
filtered = rescaled
|
35 |
-
# filtered[filtered[:, :, 0] < min_th/100, 0] = 0
|
36 |
-
# filtered[filtered[:, :, 0] < min_th/100, 1] = 0
|
37 |
-
# filtered[filtered[:, :, 0] < min_th/100, 2] = 0
|
38 |
-
# filt_arr = filtered[((filtered[:,0] > min_th/100) & (filtered[:,0] < max_th/100))]
|
39 |
-
filt_arr = (filtered > min_th/100) * filtered * (filtered < max_th/100)
|
40 |
|
|
|
41 |
|
42 |
-
print("Shape Image:",image.shape)
|
43 |
-
print("Shape Image filt:",
|
44 |
-
print("Shape Image Heat:",
|
45 |
-
|
46 |
-
return image_out, im_filt, image
|
47 |
|
48 |
# def detr(im):
|
49 |
# return im
|
50 |
|
51 |
gr_input = [
|
52 |
-
gr.inputs.Image(label="
|
53 |
-
,gr.inputs.Slider(minimum=0, maximum=100, step=5, default=0, label="Minimum Threshold")
|
54 |
-
,gr.inputs.Slider(minimum=0, maximum=100, step=5, default=100, label="Maximum Threshold")
|
55 |
]
|
56 |
|
57 |
gr_output = [
|
58 |
-
gr.outputs.Image(type="pil",label="HeatMap
|
59 |
-
gr.outputs.Image(type="pil",label="Filtered Image"),
|
60 |
-
gr.outputs.Image(type="pil",label="
|
|
|
61 |
]
|
62 |
|
63 |
iface = gr.Interface(
|
64 |
fn=infer,
|
65 |
-
title="
|
66 |
-
description = "
|
67 |
inputs = gr_input,
|
68 |
-
outputs = gr_output
|
|
|
|
|
|
|
|
|
|
|
69 |
)
|
70 |
iface.launch()
|
|
|
6 |
|
7 |
from huggingface_hub import from_pretrained_keras
|
8 |
|
9 |
+
from einops import repeat
|
10 |
+
|
11 |
custom_objects = {'BilinearUpSampling2D': BilinearUpSampling2D, 'depth_loss_function': None}
|
12 |
print('Loading model...')
|
13 |
model = from_pretrained_keras("keras-io/monocular-depth-estimation", custom_objects=custom_objects, compile=False)
|
|
|
17 |
import utils
|
18 |
importlib.reload(utils)
|
19 |
|
20 |
+
def layer_over_image(raw_image, filter, custom_color = [0, 0, 0]):
|
21 |
+
# print(raw_image[:, :, 0])
|
22 |
+
out_image = raw_image
|
23 |
+
out_image[:,:,0] = raw_image[:, :, 0] * filter
|
24 |
+
out_image[:,:,1] = raw_image[:, :, 1] * filter
|
25 |
+
out_image[:,:,2] = raw_image[:, :, 2] * filter
|
26 |
+
return raw_image
|
27 |
+
|
28 |
def infer(image, min_th, max_th):
|
29 |
+
# print('_'*20)
|
30 |
inputs = utils.load_images([image])
|
31 |
outputs = utils.predict(model, inputs)
|
32 |
|
33 |
plasma = plt.get_cmap('plasma')
|
34 |
rescaled = outputs[0][:, :, 0]
|
35 |
+
# print("Min Max Bef", np.min(rescaled), np.max(rescaled))
|
36 |
rescaled = rescaled - np.min(rescaled)
|
37 |
rescaled = rescaled / np.max(rescaled)
|
38 |
|
39 |
+
im_heat = plasma(rescaled)[:, :, :3]
|
40 |
+
|
41 |
+
# print("Min Max Aft", np.min(rescaled), np.max(rescaled))
|
42 |
+
|
43 |
+
# print("Shape Scaled:",rescaled.shape)
|
44 |
+
filt_base = rescaled
|
45 |
+
filt_base = repeat(filt_base, "h w -> (h 2) (w 2)")
|
46 |
+
filt_arr_min = (filt_base > min_th/100)
|
47 |
+
filt_arr_max = (filt_base < max_th/100)
|
48 |
+
|
49 |
+
filt_arr = filt_arr_min * filt_base * filt_arr_max
|
50 |
+
im_heat_filt = plasma(filt_arr)[:, :, :3]
|
51 |
|
52 |
+
if max_th < 100:
|
53 |
+
image_emph = layer_over_image(image, filt_arr_max)
|
54 |
+
else:
|
55 |
+
image_emph = image
|
56 |
+
|
57 |
+
if min_th > 0:
|
58 |
+
image_emph = layer_over_image(image, filt_arr_min)
|
59 |
+
else:
|
60 |
+
image_emph = image
|
61 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
|
63 |
+
# print("filt arr min", filt_arr_min)
|
64 |
|
65 |
+
# print("Shape Image:",image.shape)
|
66 |
+
# print("Shape Image filt:",im_heat_filt.shape)
|
67 |
+
# print("Shape Image Heat:",im_heat.shape)
|
68 |
+
return im_heat, image_emph
|
|
|
69 |
|
70 |
# def detr(im):
|
71 |
# return im
|
72 |
|
73 |
gr_input = [
|
74 |
+
gr.inputs.Image(label="Image", type="numpy", shape=(640, 480))
|
75 |
+
,gr.inputs.Slider(minimum=0, maximum=100, step=0.5, default=0, label="Minimum Threshold")
|
76 |
+
,gr.inputs.Slider(minimum=0, maximum=100, step=0.5, default=100, label="Maximum Threshold")
|
77 |
]
|
78 |
|
79 |
gr_output = [
|
80 |
+
gr.outputs.Image(type="pil",label="Depth HeatMap"),
|
81 |
+
# gr.outputs.Image(type="pil",label="Filtered Image"),
|
82 |
+
# gr.outputs.Image(type="pil",label="Before"),
|
83 |
+
gr.outputs.Image(type="pil",label="Important Areas")
|
84 |
]
|
85 |
|
86 |
iface = gr.Interface(
|
87 |
fn=infer,
|
88 |
+
title="Monocular Depth Filter",
|
89 |
+
description = "Used Keras Depth Estimation Model for estimating the depth of areas in an image. Image is then filtered out to only show the selected areas",
|
90 |
inputs = gr_input,
|
91 |
+
outputs = gr_output,
|
92 |
+
examples=[
|
93 |
+
["examples/00015_colors.png", 0, 90]
|
94 |
+
,["examples/car.jpg", 0, 30]
|
95 |
+
,["examples/dash.jpg", 10, 55]
|
96 |
+
]
|
97 |
)
|
98 |
iface.launch()
|
examples/00015_colors.png
ADDED
![]() |
examples/car.JPG
ADDED
|
examples/dash.jpg
ADDED
![]() |
utils.py
CHANGED
@@ -12,8 +12,8 @@ def predict(model, images, minDepth=10, maxDepth=1000, batch_size=2):
|
|
12 |
# Compute predictions
|
13 |
predictions = model.predict(images, batch_size=batch_size)
|
14 |
# Put in expected range
|
15 |
-
print("Max Depth:", np.amax(predictions), maxDepth)
|
16 |
-
print("Min Depth:", np.amin(predictions), minDepth)
|
17 |
return np.clip(depth_norm(predictions, maxDepth=maxDepth), minDepth, maxDepth) / maxDepth
|
18 |
|
19 |
|
|
|
12 |
# Compute predictions
|
13 |
predictions = model.predict(images, batch_size=batch_size)
|
14 |
# Put in expected range
|
15 |
+
# print("Max Depth:", np.amax(predictions), maxDepth)
|
16 |
+
# print("Min Depth:", np.amin(predictions), minDepth)
|
17 |
return np.clip(depth_norm(predictions, maxDepth=maxDepth), minDepth, maxDepth) / maxDepth
|
18 |
|
19 |
|