Update app.py
Browse files
app.py
CHANGED
@@ -116,22 +116,44 @@ sample_images = get_sample_images('example_images')
|
|
116 |
gr.Interface(
|
117 |
predict,
|
118 |
title='Land Cover Segmentation',
|
119 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
120 |
'''
|
121 |
The DeepGlobe Land Cover Classification Challenge offers the first public dataset containing high resolution
|
122 |
satellite imagery focusing on rural areas. As there are multiple land cover types and high density of annotations,
|
123 |
this dataset is more challenging than its counterparts launched in 2018. All satellite images contain RGB pixels,
|
124 |
with a pixel resolution of 50 cm. The total size of the total area of the dataset is equivalent to 10716.9 square kilometers.
|
125 |
|
126 |
-
|
127 |
-
I trained on 803 images and their segmentation masks (with split of 80/20%). For this multilabel segmentation task, we trained 4 models,
|
128 |
the basic 4-blocks U-net CNN, VGG16 U-Net, Resnet50 U-net and Efficient Net U-net. Then, I built an ensemble model that achieved a
|
129 |
validation accuracy of about 75% and dice score of about 0.6.
|
130 |
-
'''
|
131 |
-
inputs=[gr.Image()],
|
132 |
-
outputs=[gr.Image()],
|
133 |
-
examples=sample_images
|
134 |
-
).launch(debug=True, share=True)
|
135 |
|
136 |
# Launch the interface
|
137 |
iface.launch(share=True)
|
|
|
116 |
gr.Interface(
|
117 |
predict,
|
118 |
title='Land Cover Segmentation',
|
119 |
+
inputs=[gr.Image()],
|
120 |
+
outputs=[gr.Image()],
|
121 |
+
examples=sample_images
|
122 |
+
).launch(debug=True, share=True)
|
123 |
+
|
124 |
+
tab1 = gr.Interface(
|
125 |
+
fn=predict,
|
126 |
+
inputs=gr.Image(label='', type="pil"),
|
127 |
+
outputs=[gr.Image(type="pil"), gr.Image(type="pil")],
|
128 |
+
title='Images with Ground Truth',
|
129 |
+
examples=sample_images,
|
130 |
+
tab="Train"
|
131 |
+
)
|
132 |
+
|
133 |
+
# Create the video processing interface
|
134 |
+
tab2 = gr.Interface(
|
135 |
+
fn=process_video,
|
136 |
+
inputs=gr.File(label=""),
|
137 |
+
outputs=gr.File(label=""),
|
138 |
+
title='Images with Ground Truth',
|
139 |
+
examples=sample_images,
|
140 |
+
tab="Test"
|
141 |
+
)
|
142 |
+
|
143 |
+
# Create a Multi Interface with Tabs
|
144 |
+
iface = gr.TabbedInterface([tab1, tab2],
|
145 |
+
title='Land Cover Segmentation',
|
146 |
+
description=
|
147 |
'''
|
148 |
The DeepGlobe Land Cover Classification Challenge offers the first public dataset containing high resolution
|
149 |
satellite imagery focusing on rural areas. As there are multiple land cover types and high density of annotations,
|
150 |
this dataset is more challenging than its counterparts launched in 2018. All satellite images contain RGB pixels,
|
151 |
with a pixel resolution of 50 cm. The total size of the total area of the dataset is equivalent to 10716.9 square kilometers.
|
152 |
|
153 |
+
We trained on 803 images and their segmentation masks (with split of 80/20%). For this multilabel segmentation task, we trained 4 models,
|
|
|
154 |
the basic 4-blocks U-net CNN, VGG16 U-Net, Resnet50 U-net and Efficient Net U-net. Then, I built an ensemble model that achieved a
|
155 |
validation accuracy of about 75% and dice score of about 0.6.
|
156 |
+
''')
|
|
|
|
|
|
|
|
|
157 |
|
158 |
# Launch the interface
|
159 |
iface.launch(share=True)
|