fix bugs
Browse files
app.py
CHANGED
@@ -28,8 +28,11 @@ def process_image(image, prompt):
|
|
28 |
outputs = model(**inputs)
|
29 |
preds = outputs.logits
|
30 |
|
31 |
-
filename =
|
32 |
-
|
|
|
|
|
|
|
33 |
return Image.open("mask.png").convert("RGB")
|
34 |
|
35 |
|
@@ -39,8 +42,8 @@ def read_content(file_path):
|
|
39 |
return content
|
40 |
|
41 |
|
42 |
-
def predict(
|
43 |
-
width, height =
|
44 |
if width < height:
|
45 |
factor = width / 512.0
|
46 |
width = 512
|
@@ -51,10 +54,10 @@ def predict(dict, text_query, reference, scale, seed, step):
|
|
51 |
height = 512
|
52 |
width = int((width / factor) / 8.0) * 8
|
53 |
|
54 |
-
init_image =
|
55 |
-
mask =
|
56 |
-
|
57 |
-
|
58 |
generator = torch.Generator('cuda').manual_seed(seed) if seed != 0 else None
|
59 |
output = pipe(
|
60 |
image=init_image,
|
@@ -122,7 +125,7 @@ with image_blocks as demo:
|
|
122 |
with gr.Box():
|
123 |
with gr.Row():
|
124 |
with gr.Column():
|
125 |
-
image = gr.Image(source="upload",
|
126 |
text = gr.Textbox(lines=1, placeholder="Clothing item you want to replace...")
|
127 |
reference = gr.Image(source="upload", elem_id="image_upload", type="pil", label="Reference Image")
|
128 |
|
@@ -146,11 +149,25 @@ with image_blocks as demo:
|
|
146 |
|
147 |
with gr.Row():
|
148 |
with gr.Column():
|
149 |
-
gr.Examples(
|
|
|
|
|
|
|
|
|
|
|
150 |
with gr.Column():
|
151 |
-
gr.Examples(
|
|
|
|
|
|
|
|
|
|
|
152 |
|
153 |
-
btn.click(
|
|
|
|
|
|
|
|
|
154 |
share_button.click(None, [], [], _js=share_js)
|
155 |
|
156 |
gr.HTML(
|
|
|
28 |
outputs = model(**inputs)
|
29 |
preds = outputs.logits
|
30 |
|
31 |
+
filename = "mask.png"
|
32 |
+
preds = torch.sigmoid(preds)
|
33 |
+
preds[preds >= 0.5] = 1
|
34 |
+
preds[preds < 0.5] = 0
|
35 |
+
plt.imsave(filename, preds)
|
36 |
return Image.open("mask.png").convert("RGB")
|
37 |
|
38 |
|
|
|
42 |
return content
|
43 |
|
44 |
|
45 |
+
def predict(input_image, text_query, reference, scale, seed, step):
|
46 |
+
width, height = input_image.size
|
47 |
if width < height:
|
48 |
factor = width / 512.0
|
49 |
width = 512
|
|
|
54 |
height = 512
|
55 |
width = int((width / factor) / 8.0) * 8
|
56 |
|
57 |
+
init_image = input_image.convert("RGB").resize((width, height))
|
58 |
+
mask = process_image(input_image, text_query).resize((width, height))
|
59 |
+
#mask = dict["mask"].convert("RGB").resize((width, height))
|
60 |
+
|
61 |
generator = torch.Generator('cuda').manual_seed(seed) if seed != 0 else None
|
62 |
output = pipe(
|
63 |
image=init_image,
|
|
|
125 |
with gr.Box():
|
126 |
with gr.Row():
|
127 |
with gr.Column():
|
128 |
+
image = gr.Image(source="upload", elem_id="image_upload", type="pil", label="Source Image")
|
129 |
text = gr.Textbox(lines=1, placeholder="Clothing item you want to replace...")
|
130 |
reference = gr.Image(source="upload", elem_id="image_upload", type="pil", label="Reference Image")
|
131 |
|
|
|
149 |
|
150 |
with gr.Row():
|
151 |
with gr.Column():
|
152 |
+
gr.Examples(
|
153 |
+
image_list,
|
154 |
+
inputs=[image],
|
155 |
+
label="Examples - Source Image",
|
156 |
+
examples_per_page=12
|
157 |
+
)
|
158 |
with gr.Column():
|
159 |
+
gr.Examples(
|
160 |
+
ref_list,
|
161 |
+
inputs=[reference],
|
162 |
+
label="Examples - Reference Image",
|
163 |
+
examples_per_page=12
|
164 |
+
)
|
165 |
|
166 |
+
btn.click(
|
167 |
+
fn=predict,
|
168 |
+
inputs=[image, text, reference, guidance, seed, steps],
|
169 |
+
outputs=[image_out, community_icon, loading_icon, share_button]
|
170 |
+
)
|
171 |
share_button.click(None, [], [], _js=share_js)
|
172 |
|
173 |
gr.HTML(
|