fix UI for gradio upgrade
Browse files
app.py
CHANGED
@@ -31,8 +31,14 @@ embedding_powers = [1. for i in range(max_tabs)]
|
|
31 |
embedding_base64s = [None for i in range(max_tabs)]
|
32 |
# embedding_base64s = gr.State(value=[None for i in range(max_tabs)])
|
33 |
|
|
|
|
|
|
|
|
|
|
|
34 |
|
35 |
def image_to_embedding(input_im):
|
|
|
36 |
input_im = Image.fromarray(input_im)
|
37 |
prepro = preprocess(input_im).unsqueeze(0).to(device)
|
38 |
with torch.no_grad():
|
@@ -42,6 +48,7 @@ def image_to_embedding(input_im):
|
|
42 |
return image_embeddings_np
|
43 |
|
44 |
def prompt_to_embedding(prompt):
|
|
|
45 |
text = tokenizer([prompt]).to(device)
|
46 |
with torch.no_grad():
|
47 |
prompt_embededdings = model.encode_text(text)
|
@@ -50,6 +57,7 @@ def prompt_to_embedding(prompt):
|
|
50 |
return prompt_embededdings_np
|
51 |
|
52 |
def embedding_to_image(embeddings):
|
|
|
53 |
size = math.ceil(math.sqrt(embeddings.shape[0]))
|
54 |
image_embeddings_square = np.pad(embeddings, (0, size**2 - embeddings.shape[0]), 'constant')
|
55 |
image_embeddings_square.resize(size,size)
|
@@ -57,6 +65,7 @@ def embedding_to_image(embeddings):
|
|
57 |
return embedding_image
|
58 |
|
59 |
def embedding_to_base64(embeddings):
|
|
|
60 |
import base64
|
61 |
# ensure float32
|
62 |
embeddings = embeddings.astype(np.float32)
|
@@ -64,12 +73,22 @@ def embedding_to_base64(embeddings):
|
|
64 |
return embeddings_b64
|
65 |
|
66 |
def base64_to_embedding(embeddings_b64):
|
|
|
67 |
import base64
|
68 |
embeddings = base64.urlsafe_b64decode(embeddings_b64)
|
69 |
embeddings = np.frombuffer(embeddings, dtype=np.float32)
|
70 |
# embeddings = torch.tensor(embeddings)
|
71 |
return embeddings
|
72 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
def safe_url(url):
|
74 |
import urllib.parse
|
75 |
url = urllib.parse.quote(url, safe=':/')
|
@@ -83,6 +102,7 @@ def main(
|
|
83 |
embeddings,
|
84 |
n_samples=4,
|
85 |
):
|
|
|
86 |
|
87 |
embeddings = base64_to_embedding(embeddings)
|
88 |
# convert to python array
|
@@ -117,17 +137,21 @@ def main(
|
|
117 |
return images
|
118 |
|
119 |
def on_image_load_update_embeddings(image_data):
|
|
|
120 |
# image to embeddings
|
121 |
if image_data is None:
|
122 |
# embeddings = prompt_to_embedding('')
|
123 |
# embeddings_b64 = embedding_to_base64(embeddings)
|
124 |
# return gr.Text.update(embeddings_b64)
|
125 |
-
return gr.Text.update('')
|
|
|
126 |
embeddings = image_to_embedding(image_data)
|
127 |
embeddings_b64 = embedding_to_base64(embeddings)
|
128 |
-
return gr.Text.update(embeddings_b64)
|
|
|
129 |
|
130 |
def on_prompt_change_update_embeddings(prompt):
|
|
|
131 |
# prompt to embeddings
|
132 |
if prompt is None or prompt == "":
|
133 |
embeddings = prompt_to_embedding('')
|
@@ -135,9 +159,10 @@ def on_prompt_change_update_embeddings(prompt):
|
|
135 |
return gr.Text.update(embedding_to_base64(embeddings))
|
136 |
embeddings = prompt_to_embedding(prompt)
|
137 |
embeddings_b64 = embedding_to_base64(embeddings)
|
138 |
-
return
|
139 |
|
140 |
def update_average_embeddings(embedding_base64s_state, embedding_powers):
|
|
|
141 |
final_embedding = None
|
142 |
num_embeddings = 0
|
143 |
for i, embedding_base64 in enumerate(embedding_base64s_state):
|
@@ -154,7 +179,7 @@ def update_average_embeddings(embedding_base64s_state, embedding_powers):
|
|
154 |
# embeddings = prompt_to_embedding('')
|
155 |
# embeddings_b64 = embedding_to_base64(embeddings)
|
156 |
# return gr.Text.update(embeddings_b64)
|
157 |
-
return
|
158 |
|
159 |
# TODO toggle this to support average or sum
|
160 |
# final_embedding = final_embedding / num_embeddings
|
@@ -166,22 +191,25 @@ def update_average_embeddings(embedding_base64s_state, embedding_powers):
|
|
166 |
return embeddings_b64
|
167 |
|
168 |
def on_power_change_update_average_embeddings(embedding_base64s_state, embedding_power_state, power, idx):
|
|
|
169 |
embedding_power_state[idx] = power
|
170 |
embeddings_b64 = update_average_embeddings(embedding_base64s_state, embedding_power_state)
|
171 |
-
return
|
172 |
|
173 |
def on_embeddings_changed_update_average_embeddings(embedding_base64s_state, embedding_power_state, embedding_base64, idx):
|
|
|
174 |
embedding_base64s_state[idx] = embedding_base64 if embedding_base64 != '' else None
|
175 |
embeddings_b64 = update_average_embeddings(embedding_base64s_state, embedding_power_state)
|
176 |
-
return
|
177 |
|
178 |
def on_embeddings_changed_update_plot(embeddings_b64):
|
|
|
179 |
# plot new embeddings
|
180 |
if embeddings_b64 is None or embeddings_b64 == "":
|
181 |
data = pd.DataFrame({
|
182 |
'embedding': [],
|
183 |
'index': []})
|
184 |
-
|
185 |
x="index",
|
186 |
y="embedding",
|
187 |
# color="country",
|
@@ -192,6 +220,7 @@ def on_embeddings_changed_update_plot(embeddings_b64):
|
|
192 |
# stroke_dash_legend_title="Country Cluster",
|
193 |
# height=300,
|
194 |
width=0)
|
|
|
195 |
|
196 |
embeddings = base64_to_embedding(embeddings_b64)
|
197 |
data = pd.DataFrame({
|
@@ -210,6 +239,7 @@ def on_embeddings_changed_update_plot(embeddings_b64):
|
|
210 |
width=embeddings.shape[0])
|
211 |
|
212 |
def on_example_image_click_set_image(input_image, image_url):
|
|
|
213 |
input_image.value = image_url
|
214 |
|
215 |
# device = torch.device("mps" if torch.backends.mps.is_available() else "cuda:0" if torch.cuda.is_available() else "cpu")
|
@@ -236,7 +266,7 @@ examples = [
|
|
236 |
# ["SohoJoeEth.jpeg", "Snoop Dogg.jpg", "SohoJoeEth + Snoop Dogg.jpeg"],
|
237 |
["pup1.jpg", "", "Pup no teacup.jpg"],
|
238 |
]
|
239 |
-
tile_size =
|
240 |
# image_folder = os.path.join("file", "images")
|
241 |
image_folder ="images"
|
242 |
|
@@ -349,7 +379,7 @@ Try uploading a few images and/or add some text prompts and search the embedding
|
|
349 |
# input_image.change(on_image_load, inputs= [input_image, plot])
|
350 |
with gr.Row():
|
351 |
with gr.Column(scale=2, min_width=240):
|
352 |
-
input_prompts[i] = gr.Textbox(label="Text Prompt", show_label=True)
|
353 |
with gr.Column(scale=3, min_width=600):
|
354 |
with gr.Row():
|
355 |
# with gr.Slider(min=-5, max=5, value=1, label="Power", show_label=True):
|
@@ -357,7 +387,7 @@ Try uploading a few images and/or add some text prompts and search the embedding
|
|
357 |
embedding_powers[i] = gr.Slider(minimum=-3, maximum=3, value=1, label="Power", show_label=True, interactive=True)
|
358 |
with gr.Row():
|
359 |
with gr.Accordion(f"Embeddings (base64)", open=False):
|
360 |
-
embedding_base64s[i] = gr.Textbox(show_label=False)
|
361 |
for idx, (tab_title, examples) in enumerate(tabbed_examples.items()):
|
362 |
with gr.Tab(tab_title):
|
363 |
with gr.Row():
|
@@ -395,15 +425,52 @@ Try uploading a few images and/or add some text prompts and search the embedding
|
|
395 |
|
396 |
embedding_base64s_state = gr.State(value=[None for i in range(max_tabs)])
|
397 |
embedding_power_state = gr.State(value=[1. for i in range(max_tabs)])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
398 |
for i in range(max_tabs):
|
399 |
-
input_images[i].change(on_image_load_update_embeddings, input_images[i], [embedding_base64s[i]])
|
400 |
-
input_prompts[i].change(on_prompt_change_update_embeddings, input_prompts[i], [embedding_base64s[i]])
|
401 |
-
embedding_base64s[i].change(on_embeddings_changed_update_plot, embedding_base64s[i], [embedding_plots[i]])
|
402 |
idx_state = gr.State(value=i)
|
403 |
-
|
404 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
405 |
|
406 |
-
average_embedding_base64.change(on_embeddings_changed_update_plot, average_embedding_base64, average_embedding_plot)
|
407 |
|
408 |
# submit.click(main, inputs= [embedding_base64s[0], scale, n_samples, steps, seed], outputs=output)
|
409 |
submit.click(main, inputs= [average_embedding_base64, n_samples], outputs=output)
|
@@ -439,4 +506,4 @@ My interest is to use CLIP for image/video understanding (see [CLIP_visual-spati
|
|
439 |
# ![Alt Text](file/pup1.jpg){height=100 width=100}
|
440 |
|
441 |
if __name__ == "__main__":
|
442 |
-
demo.launch()
|
|
|
31 |
embedding_base64s = [None for i in range(max_tabs)]
|
32 |
# embedding_base64s = gr.State(value=[None for i in range(max_tabs)])
|
33 |
|
34 |
+
debug_print_on = False
|
35 |
+
|
36 |
+
def debug_print(*args, **kwargs):
|
37 |
+
if debug_print_on:
|
38 |
+
print(*args, **kwargs)
|
39 |
|
40 |
def image_to_embedding(input_im):
|
41 |
+
# debug_print("image_to_embedding")
|
42 |
input_im = Image.fromarray(input_im)
|
43 |
prepro = preprocess(input_im).unsqueeze(0).to(device)
|
44 |
with torch.no_grad():
|
|
|
48 |
return image_embeddings_np
|
49 |
|
50 |
def prompt_to_embedding(prompt):
|
51 |
+
# debug_print("prompt_to_embedding")
|
52 |
text = tokenizer([prompt]).to(device)
|
53 |
with torch.no_grad():
|
54 |
prompt_embededdings = model.encode_text(text)
|
|
|
57 |
return prompt_embededdings_np
|
58 |
|
59 |
def embedding_to_image(embeddings):
|
60 |
+
# debug_print("embedding_to_image")
|
61 |
size = math.ceil(math.sqrt(embeddings.shape[0]))
|
62 |
image_embeddings_square = np.pad(embeddings, (0, size**2 - embeddings.shape[0]), 'constant')
|
63 |
image_embeddings_square.resize(size,size)
|
|
|
65 |
return embedding_image
|
66 |
|
67 |
def embedding_to_base64(embeddings):
|
68 |
+
# debug_print("embedding_to_base64")
|
69 |
import base64
|
70 |
# ensure float32
|
71 |
embeddings = embeddings.astype(np.float32)
|
|
|
73 |
return embeddings_b64
|
74 |
|
75 |
def base64_to_embedding(embeddings_b64):
|
76 |
+
# debug_print("base64_to_embedding")
|
77 |
import base64
|
78 |
embeddings = base64.urlsafe_b64decode(embeddings_b64)
|
79 |
embeddings = np.frombuffer(embeddings, dtype=np.float32)
|
80 |
# embeddings = torch.tensor(embeddings)
|
81 |
return embeddings
|
82 |
|
83 |
+
def is_prompt_embeddings(prompt):
|
84 |
+
if prompt is None or prompt == "":
|
85 |
+
return False
|
86 |
+
try:
|
87 |
+
embedding = base64_to_embedding(prompt)
|
88 |
+
return True
|
89 |
+
except Exception as e:
|
90 |
+
return False
|
91 |
+
|
92 |
def safe_url(url):
|
93 |
import urllib.parse
|
94 |
url = urllib.parse.quote(url, safe=':/')
|
|
|
102 |
embeddings,
|
103 |
n_samples=4,
|
104 |
):
|
105 |
+
debug_print("main")
|
106 |
|
107 |
embeddings = base64_to_embedding(embeddings)
|
108 |
# convert to python array
|
|
|
137 |
return images
|
138 |
|
139 |
def on_image_load_update_embeddings(image_data):
|
140 |
+
debug_print("on_image_load_update_embeddings")
|
141 |
# image to embeddings
|
142 |
if image_data is None:
|
143 |
# embeddings = prompt_to_embedding('')
|
144 |
# embeddings_b64 = embedding_to_base64(embeddings)
|
145 |
# return gr.Text.update(embeddings_b64)
|
146 |
+
# return gr.Text.update('')
|
147 |
+
return ''
|
148 |
embeddings = image_to_embedding(image_data)
|
149 |
embeddings_b64 = embedding_to_base64(embeddings)
|
150 |
+
# return gr.Text.update(embeddings_b64)
|
151 |
+
return embeddings_b64
|
152 |
|
153 |
def on_prompt_change_update_embeddings(prompt):
|
154 |
+
debug_print("on_prompt_change_update_embeddings")
|
155 |
# prompt to embeddings
|
156 |
if prompt is None or prompt == "":
|
157 |
embeddings = prompt_to_embedding('')
|
|
|
159 |
return gr.Text.update(embedding_to_base64(embeddings))
|
160 |
embeddings = prompt_to_embedding(prompt)
|
161 |
embeddings_b64 = embedding_to_base64(embeddings)
|
162 |
+
return embeddings_b64
|
163 |
|
164 |
def update_average_embeddings(embedding_base64s_state, embedding_powers):
|
165 |
+
debug_print("update_average_embeddings")
|
166 |
final_embedding = None
|
167 |
num_embeddings = 0
|
168 |
for i, embedding_base64 in enumerate(embedding_base64s_state):
|
|
|
179 |
# embeddings = prompt_to_embedding('')
|
180 |
# embeddings_b64 = embedding_to_base64(embeddings)
|
181 |
# return gr.Text.update(embeddings_b64)
|
182 |
+
return ''
|
183 |
|
184 |
# TODO toggle this to support average or sum
|
185 |
# final_embedding = final_embedding / num_embeddings
|
|
|
191 |
return embeddings_b64
|
192 |
|
193 |
def on_power_change_update_average_embeddings(embedding_base64s_state, embedding_power_state, power, idx):
|
194 |
+
debug_print("on_power_change_update_average_embeddings")
|
195 |
embedding_power_state[idx] = power
|
196 |
embeddings_b64 = update_average_embeddings(embedding_base64s_state, embedding_power_state)
|
197 |
+
return embeddings_b64
|
198 |
|
199 |
def on_embeddings_changed_update_average_embeddings(embedding_base64s_state, embedding_power_state, embedding_base64, idx):
|
200 |
+
debug_print("on_embeddings_changed_update_average_embeddings")
|
201 |
embedding_base64s_state[idx] = embedding_base64 if embedding_base64 != '' else None
|
202 |
embeddings_b64 = update_average_embeddings(embedding_base64s_state, embedding_power_state)
|
203 |
+
return embeddings_b64
|
204 |
|
205 |
def on_embeddings_changed_update_plot(embeddings_b64):
|
206 |
+
debug_print("on_embeddings_changed_update_plot")
|
207 |
# plot new embeddings
|
208 |
if embeddings_b64 is None or embeddings_b64 == "":
|
209 |
data = pd.DataFrame({
|
210 |
'embedding': [],
|
211 |
'index': []})
|
212 |
+
update = gr.LinePlot.update(data,
|
213 |
x="index",
|
214 |
y="embedding",
|
215 |
# color="country",
|
|
|
220 |
# stroke_dash_legend_title="Country Cluster",
|
221 |
# height=300,
|
222 |
width=0)
|
223 |
+
return update
|
224 |
|
225 |
embeddings = base64_to_embedding(embeddings_b64)
|
226 |
data = pd.DataFrame({
|
|
|
239 |
width=embeddings.shape[0])
|
240 |
|
241 |
def on_example_image_click_set_image(input_image, image_url):
|
242 |
+
debug_print("on_example_image_click_set_image")
|
243 |
input_image.value = image_url
|
244 |
|
245 |
# device = torch.device("mps" if torch.backends.mps.is_available() else "cuda:0" if torch.cuda.is_available() else "cpu")
|
|
|
266 |
# ["SohoJoeEth.jpeg", "Snoop Dogg.jpg", "SohoJoeEth + Snoop Dogg.jpeg"],
|
267 |
["pup1.jpg", "", "Pup no teacup.jpg"],
|
268 |
]
|
269 |
+
tile_size = 110
|
270 |
# image_folder = os.path.join("file", "images")
|
271 |
image_folder ="images"
|
272 |
|
|
|
379 |
# input_image.change(on_image_load, inputs= [input_image, plot])
|
380 |
with gr.Row():
|
381 |
with gr.Column(scale=2, min_width=240):
|
382 |
+
input_prompts[i] = gr.Textbox(label="Text Prompt", show_label=True, max_lines=4)
|
383 |
with gr.Column(scale=3, min_width=600):
|
384 |
with gr.Row():
|
385 |
# with gr.Slider(min=-5, max=5, value=1, label="Power", show_label=True):
|
|
|
387 |
embedding_powers[i] = gr.Slider(minimum=-3, maximum=3, value=1, label="Power", show_label=True, interactive=True)
|
388 |
with gr.Row():
|
389 |
with gr.Accordion(f"Embeddings (base64)", open=False):
|
390 |
+
embedding_base64s[i] = gr.Textbox(show_label=False, live=True)
|
391 |
for idx, (tab_title, examples) in enumerate(tabbed_examples.items()):
|
392 |
with gr.Tab(tab_title):
|
393 |
with gr.Row():
|
|
|
425 |
|
426 |
embedding_base64s_state = gr.State(value=[None for i in range(max_tabs)])
|
427 |
embedding_power_state = gr.State(value=[1. for i in range(max_tabs)])
|
428 |
+
|
429 |
+
def on_image_load(input_image, idx_state, embedding_base64s_state, embedding_power_state):
|
430 |
+
debug_print("on_image_load")
|
431 |
+
embeddings_b64 = on_image_load_update_embeddings(input_image)
|
432 |
+
new_plot = on_embeddings_changed_update_plot(embeddings_b64)
|
433 |
+
average_embeddings_b64 = on_embeddings_changed_update_average_embeddings(embedding_base64s_state, embedding_power_state, embeddings_b64, idx_state)
|
434 |
+
new_average_plot = on_embeddings_changed_update_plot(average_embeddings_b64)
|
435 |
+
return embeddings_b64, new_plot, average_embeddings_b64, new_average_plot
|
436 |
+
|
437 |
+
def on_prompt_change(prompt, idx_state, embedding_base64s_state, embedding_power_state):
|
438 |
+
debug_print("on_prompt_change")
|
439 |
+
if is_prompt_embeddings(prompt):
|
440 |
+
embeddings_b64 = prompt
|
441 |
+
else:
|
442 |
+
embeddings_b64 = on_prompt_change_update_embeddings(prompt)
|
443 |
+
new_plot = on_embeddings_changed_update_plot(embeddings_b64)
|
444 |
+
average_embeddings_b64 = on_embeddings_changed_update_average_embeddings(embedding_base64s_state, embedding_power_state, embeddings_b64, idx_state)
|
445 |
+
new_average_plot = on_embeddings_changed_update_plot(average_embeddings_b64)
|
446 |
+
return embeddings_b64, new_plot, average_embeddings_b64, new_average_plot
|
447 |
+
|
448 |
+
def on_power_change(power, idx_state, embedding_base64s_state, embedding_power_state):
|
449 |
+
debug_print("on_power_change")
|
450 |
+
average_embeddings_b64 = on_power_change_update_average_embeddings(embedding_base64s_state, embedding_power_state, power, idx_state)
|
451 |
+
new_average_plot = on_embeddings_changed_update_plot(average_embeddings_b64)
|
452 |
+
return average_embeddings_b64, new_average_plot
|
453 |
+
|
454 |
for i in range(max_tabs):
|
|
|
|
|
|
|
455 |
idx_state = gr.State(value=i)
|
456 |
+
input_images[i].change(on_image_load,
|
457 |
+
[input_images[i], idx_state, embedding_base64s_state, embedding_power_state],
|
458 |
+
[embedding_base64s[i], embedding_plots[i], average_embedding_base64, average_embedding_plot])
|
459 |
+
input_prompts[i].change(on_prompt_change,
|
460 |
+
[input_prompts[i], idx_state, embedding_base64s_state, embedding_power_state],
|
461 |
+
[embedding_base64s[i], embedding_plots[i], average_embedding_base64, average_embedding_plot])
|
462 |
+
embedding_powers[i].change(on_power_change,
|
463 |
+
[embedding_powers[i], idx_state, embedding_base64s_state, embedding_power_state],
|
464 |
+
[average_embedding_base64, average_embedding_plot])
|
465 |
+
|
466 |
+
|
467 |
+
# input_images[i].change(on_image_load_update_embeddings, input_images[i], embedding_base64s[i])
|
468 |
+
# input_prompts[i].change(on_prompt_change_update_embeddings, input_prompts[i], embedding_base64s[i])
|
469 |
+
# embedding_base64s[i].change(on_embeddings_changed_update_plot, embedding_base64s[i], embedding_plots[i])
|
470 |
+
# embedding_base64s[i].change(on_embeddings_changed_update_average_embeddings, [embedding_base64s_state, embedding_power_state, embedding_base64s[i], idx_state], average_embedding_base64)
|
471 |
+
# embedding_powers[i].change(on_power_change_update_average_embeddings, [embedding_base64s_state, embedding_power_state, embedding_powers[i], idx_state], average_embedding_base64)
|
472 |
|
473 |
+
# average_embedding_base64.change(on_embeddings_changed_update_plot, average_embedding_base64, average_embedding_plot)
|
474 |
|
475 |
# submit.click(main, inputs= [embedding_base64s[0], scale, n_samples, steps, seed], outputs=output)
|
476 |
submit.click(main, inputs= [average_embedding_base64, n_samples], outputs=output)
|
|
|
506 |
# ![Alt Text](file/pup1.jpg){height=100 width=100}
|
507 |
|
508 |
if __name__ == "__main__":
|
509 |
+
demo.launch(debug=True)
|