Spaces:
Runtime error
Runtime error
updated app.py
Browse files
app.py
CHANGED
|
@@ -12,8 +12,49 @@ inputs = gr.inputs.Image()
|
|
| 12 |
outputs = gr.outputs.Image()
|
| 13 |
|
| 14 |
# inference fn
|
| 15 |
-
def predict(
|
| 16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
|
| 19 |
-
gr.Interface(predict, inputs = inputs, outputs = outputs)
|
|
|
|
| 12 |
outputs = gr.outputs.Image()
|
| 13 |
|
| 14 |
# inference fn
|
| 15 |
+
def predict(inputs):
|
| 16 |
+
# convert img to numpy array, resize and normalize to make the prediction
|
| 17 |
+
img = np.array(inputs)
|
| 18 |
+
|
| 19 |
+
im = tf.image.resize(img, (128, 128))
|
| 20 |
+
im = tf.cast(im, tf.float32) / 255.0
|
| 21 |
+
pred_mask = model.predict(im[tf.newaxis, ...])
|
| 22 |
+
|
| 23 |
+
# take the best performing class for each pixel
|
| 24 |
+
# the output of argmax looks like this [[1, 2, 0], ...]
|
| 25 |
+
pred_mask_arg = tf.argmax(pred_mask, axis=-1)
|
| 26 |
+
|
| 27 |
+
labels = []
|
| 28 |
+
|
| 29 |
+
# convert the prediction mask into binary masks for each class
|
| 30 |
+
binary_masks = {}
|
| 31 |
+
mask_codes = {}
|
| 32 |
+
|
| 33 |
+
# when we take tf.argmax() over pred_mask, it becomes a tensor object
|
| 34 |
+
# the shape becomes TensorShape object, looking like this TensorShape([128])
|
| 35 |
+
# we need to take get shape, convert to list and take the best one
|
| 36 |
+
|
| 37 |
+
rows = pred_mask_arg[0][1].get_shape().as_list()[0]
|
| 38 |
+
cols = pred_mask_arg[0][2].get_shape().as_list()[0]
|
| 39 |
|
| 40 |
+
for cls in range(pred_mask.shape[-1]):
|
| 41 |
+
|
| 42 |
+
binary_masks[f"mask_{cls}"] = np.zeros(shape = (pred_mask.shape[1], pred_mask.shape[2])) #create masks for each class
|
| 43 |
+
|
| 44 |
+
for row in range(rows):
|
| 45 |
+
|
| 46 |
+
for col in range(cols):
|
| 47 |
+
|
| 48 |
+
if pred_mask_arg[0][row][col] == cls:
|
| 49 |
+
|
| 50 |
+
binary_masks[f"mask_{cls}"][row][col] = 1
|
| 51 |
+
else:
|
| 52 |
+
binary_masks[f"mask_{cls}"][row][col] = 0
|
| 53 |
+
|
| 54 |
+
mask = binary_masks[f"mask_{cls}"]
|
| 55 |
+
mask *= 255
|
| 56 |
+
img = Image.fromarray(mask.astype(np.int8), mode="L")
|
| 57 |
+
return img
|
| 58 |
+
|
| 59 |
|
| 60 |
+
gr.Interface(predict, inputs = inputs, outputs = outputs).launch()
|