Update app.py
Browse files
app.py
CHANGED
@@ -147,6 +147,18 @@ def predict_on_test(image):
|
|
147 |
|
148 |
sample_images = get_sample_images('example_images')
|
149 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
150 |
# Launch Gradio Interface (Single Tab interface)
|
151 |
# gr.Interface(
|
152 |
# predict,
|
@@ -161,6 +173,7 @@ tab1 = gr.Interface(
|
|
161 |
inputs=gr.Image(),
|
162 |
outputs=[gr.Image(), gr.Image()],
|
163 |
title='Images with Ground Truth',
|
|
|
164 |
examples=sample_images
|
165 |
)
|
166 |
|
@@ -170,24 +183,14 @@ tab2 = gr.Interface(
|
|
170 |
inputs=gr.Image(),
|
171 |
outputs=gr.Image(),
|
172 |
title='Images with Ground Truth',
|
|
|
173 |
examples=sample_images
|
174 |
)
|
175 |
|
176 |
# Create a Multi Interface with Tabs
|
177 |
iface = gr.TabbedInterface([tab1, tab2],
|
178 |
title='Land Cover Segmentation',
|
179 |
-
|
180 |
-
'''
|
181 |
-
The DeepGlobe Land Cover Classification Challenge offers the first public dataset containing high resolution
|
182 |
-
satellite imagery focusing on rural areas. As there are multiple land cover types and high density of annotations,
|
183 |
-
this dataset is more challenging than its counterparts launched in 2018. All satellite images contain RGB pixels,
|
184 |
-
with a pixel resolution of 50 cm. The total size of the total area of the dataset is equivalent to 10716.9 square kilometers.
|
185 |
-
|
186 |
-
We trained on 803 images and their segmentation masks (with split of 80/20%). For this multilabel segmentation task, we trained 4 models,
|
187 |
-
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
|
188 |
-
validation accuracy of about 75% and dice score of about 0.6.
|
189 |
-
''',
|
190 |
-
tab_names = ['Train','Test'])
|
191 |
|
192 |
# Launch the interface
|
193 |
iface.launch(share=True)
|
|
|
147 |
|
148 |
sample_images = get_sample_images('example_images')
|
149 |
|
150 |
+
description=
|
151 |
+
'''
|
152 |
+
The DeepGlobe Land Cover Classification Challenge offers the first public dataset containing high resolution
|
153 |
+
satellite imagery focusing on rural areas. As there are multiple land cover types and high density of annotations,
|
154 |
+
this dataset is more challenging than its counterparts launched in 2018. All satellite images contain RGB pixels,
|
155 |
+
with a pixel resolution of 50 cm. The total size of the total area of the dataset is equivalent to 10716.9 square kilometers.
|
156 |
+
|
157 |
+
We trained on 803 images and their segmentation masks (with split of 80/20%). For this multilabel segmentation task, we trained 4 models,
|
158 |
+
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
|
159 |
+
validation accuracy of about 75% and dice score of about 0.6.
|
160 |
+
'''
|
161 |
+
|
162 |
# Launch Gradio Interface (Single Tab interface)
|
163 |
# gr.Interface(
|
164 |
# predict,
|
|
|
173 |
inputs=gr.Image(),
|
174 |
outputs=[gr.Image(), gr.Image()],
|
175 |
title='Images with Ground Truth',
|
176 |
+
description=description,
|
177 |
examples=sample_images
|
178 |
)
|
179 |
|
|
|
183 |
inputs=gr.Image(),
|
184 |
outputs=gr.Image(),
|
185 |
title='Images with Ground Truth',
|
186 |
+
description=description,
|
187 |
examples=sample_images
|
188 |
)
|
189 |
|
190 |
# Create a Multi Interface with Tabs
|
191 |
iface = gr.TabbedInterface([tab1, tab2],
|
192 |
title='Land Cover Segmentation',
|
193 |
+
tab_names = ['Train','Test'])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
194 |
|
195 |
# Launch the interface
|
196 |
iface.launch(share=True)
|