Spaces:
Runtime error
Runtime error
token map v3
Browse files- app.py +71 -69
- models/region_diffusion.py +15 -28
- utils/attention_utils.py +29 -17
app.py
CHANGED
@@ -29,10 +29,10 @@ If you are encountering an error or not achieving your desired outcome, here are
|
|
29 |
|
30 |
canvas_html = """<iframe id='rich-text-root' style='width:100%' height='360px' src='file=rich-text-to-json-iframe.html' frameborder='0' scrolling='no'></iframe>"""
|
31 |
get_js_data = """
|
32 |
-
async (text_input, negative_prompt,
|
33 |
const richEl = document.getElementById("rich-text-root");
|
34 |
const data = richEl? richEl.contentDocument.body._data : {};
|
35 |
-
return [text_input, negative_prompt,
|
36 |
}
|
37 |
"""
|
38 |
set_js_data = """
|
@@ -66,27 +66,27 @@ def main():
|
|
66 |
def generate(
|
67 |
text_input: str,
|
68 |
negative_text: str,
|
69 |
-
height: int,
|
70 |
-
width: int,
|
71 |
-
seed: int,
|
72 |
-
steps: int,
|
73 |
num_segments: int,
|
74 |
segment_threshold: float,
|
75 |
inject_interval: float,
|
76 |
-
|
|
|
77 |
color_guidance_weight: float,
|
78 |
rich_text_input: str,
|
79 |
-
|
|
|
|
|
|
|
80 |
):
|
81 |
run_dir = 'results/'
|
82 |
os.makedirs(run_dir, exist_ok=True)
|
83 |
# Load region diffusion model.
|
84 |
-
height = int(height)
|
85 |
-
width = int(width)
|
86 |
steps = 41 if not steps else steps
|
87 |
guidance_weight = 8.5 if not guidance_weight else guidance_weight
|
88 |
-
text_input = rich_text_input if rich_text_input != '' else text_input
|
89 |
-
print('text_input', text_input)
|
90 |
if (text_input == '' or rich_text_input == ''):
|
91 |
raise gr.Error("Please enter some text.")
|
92 |
# parse json to span attributes
|
@@ -132,25 +132,25 @@ def main():
|
|
132 |
512//8, 512//8, region_target_token_ids[:-1], seed,
|
133 |
base_tokens, segment_threshold=segment_threshold, num_segments=num_segments,
|
134 |
return_vis=True)
|
|
|
|
|
|
|
135 |
color_obj_masks = [transforms.functional.resize(color_obj_mask, (height, width),
|
136 |
interpolation=transforms.InterpolationMode.BICUBIC,
|
137 |
antialias=True)
|
138 |
for color_obj_mask in color_obj_masks]
|
139 |
text_format_dict['color_obj_atten'] = color_obj_masks
|
|
|
140 |
model.remove_tokenmap_hooks()
|
141 |
|
142 |
# generate image from rich text
|
143 |
begin_time = time.time()
|
144 |
seed_everything(seed)
|
145 |
-
if background_aug:
|
146 |
-
bg_aug_end = 500
|
147 |
-
else:
|
148 |
-
bg_aug_end = 1000
|
149 |
rich_img = model.prompt_to_img(region_text_prompts, [negative_text],
|
150 |
height=height, width=width, num_inference_steps=steps,
|
151 |
guidance_scale=guidance_weight, use_guidance=use_grad_guidance,
|
152 |
text_format_dict=text_format_dict, inject_selfattn=inject_interval,
|
153 |
-
|
154 |
print('time lapses to generate image from rich text: %.4f' %
|
155 |
(time.time()-begin_time))
|
156 |
return [plain_img[0], rich_img[0], segments_vis, token_maps]
|
@@ -191,6 +191,12 @@ def main():
|
|
191 |
maximum=1,
|
192 |
step=0.01,
|
193 |
value=0.)
|
|
|
|
|
|
|
|
|
|
|
|
|
194 |
color_guidance_weight = gr.Slider(label='Color weight',
|
195 |
info='(To obtain more precise color, increase this, while too large value may cause artifacts.)',
|
196 |
minimum=0,
|
@@ -209,10 +215,6 @@ def main():
|
|
209 |
value=6,
|
210 |
elem_id="seed"
|
211 |
)
|
212 |
-
background_aug = gr.Checkbox(
|
213 |
-
label='Precise region alignment',
|
214 |
-
info='(For strict region alignment, select this option, but beware of potential artifacts when using with style.)',
|
215 |
-
value=True)
|
216 |
with gr.Accordion('Other Parameters', open=False):
|
217 |
steps = gr.Slider(label='Number of Steps',
|
218 |
minimum=0,
|
@@ -266,32 +268,32 @@ def main():
|
|
266 |
5,
|
267 |
0.3,
|
268 |
0,
|
|
|
269 |
6,
|
270 |
-
|
271 |
None,
|
272 |
-
True
|
273 |
],
|
274 |
[
|
275 |
-
'{"ops":[{"insert":"A "},{"attributes":{"link":"
|
276 |
'',
|
277 |
-
|
278 |
0.5,
|
279 |
0,
|
|
|
280 |
6,
|
281 |
-
|
282 |
None,
|
283 |
-
True
|
284 |
],
|
285 |
[
|
286 |
'{"ops":[{"insert":"A "},{"attributes":{"link":"Happy Kung fu panda art, elder, asian art, volumetric lighting, dramatic scene, ultra detailed, realism, chinese"},"insert":"panda"},{"insert":" standing on a cliff by a waterfall, wildlife photography, photograph, high quality, wildlife, f 1.8, soft focus, 8k, national geographic, award - winning photograph by nick nichols"}]}',
|
287 |
'',
|
288 |
-
|
289 |
0.3,
|
290 |
0,
|
|
|
291 |
4,
|
292 |
-
|
293 |
None,
|
294 |
-
True
|
295 |
],
|
296 |
]
|
297 |
|
@@ -303,10 +305,10 @@ def main():
|
|
303 |
num_segments,
|
304 |
segment_threshold,
|
305 |
inject_interval,
|
|
|
306 |
seed,
|
307 |
color_guidance_weight,
|
308 |
rich_text_input,
|
309 |
-
background_aug,
|
310 |
],
|
311 |
outputs=[
|
312 |
plaintext_result,
|
@@ -315,42 +317,42 @@ def main():
|
|
315 |
token_map,
|
316 |
],
|
317 |
fn=generate,
|
318 |
-
|
319 |
examples_per_page=20)
|
320 |
with gr.Row():
|
321 |
color_examples = [
|
322 |
[
|
323 |
-
'{"ops":[{"insert":"a beautifule girl with big eye, skin, and long "},{"attributes":{"color":"#
|
324 |
'lowres, had anatomy, bad hands, cropped, worst quality',
|
325 |
-
|
326 |
-
0.
|
|
|
327 |
0.3,
|
328 |
6,
|
329 |
0.5,
|
330 |
None,
|
331 |
-
True
|
332 |
],
|
333 |
[
|
334 |
-
'{"ops":[{"insert":"a beautifule girl with big eye, skin, and long "},{"attributes":{"color":"#
|
335 |
'lowres, had anatomy, bad hands, cropped, worst quality',
|
336 |
-
|
337 |
-
0.
|
|
|
338 |
0.3,
|
339 |
6,
|
340 |
-
0.
|
341 |
None,
|
342 |
-
True
|
343 |
],
|
344 |
[
|
345 |
'{"ops":[{"insert":"a Gothic "},{"attributes":{"color":"#FD6C9E"},"insert":"church"},{"insert":" in a the sunset with a beautiful landscape in the background."}]}',
|
346 |
'',
|
347 |
-
|
348 |
-
0.
|
349 |
0.5,
|
|
|
350 |
6,
|
351 |
0.5,
|
352 |
None,
|
353 |
-
False
|
354 |
],
|
355 |
[
|
356 |
'{"ops":[{"insert":"A mesmerizing sight that captures the beauty of a "},{"attributes":{"color":"#4775fc"},"insert":"rose"},{"insert":" blooming, close up"}]}',
|
@@ -358,21 +360,21 @@ def main():
|
|
358 |
3,
|
359 |
0.3,
|
360 |
0,
|
|
|
361 |
9,
|
362 |
1,
|
363 |
None,
|
364 |
-
False
|
365 |
],
|
366 |
[
|
367 |
'{"ops":[{"insert":"A "},{"attributes":{"color":"#FFD700"},"insert":"marble statue of a wolf\'s head and shoulder"},{"insert":", surrounded by colorful flowers michelangelo, detailed, intricate, full of color, led lighting, trending on artstation, 4 k, hyperrealistic, 3 5 mm, focused, extreme details, unreal engine 5, masterpiece "}]}',
|
368 |
'',
|
369 |
5,
|
|
|
|
|
370 |
0.3,
|
371 |
-
0,
|
372 |
5,
|
373 |
0.6,
|
374 |
None,
|
375 |
-
False
|
376 |
],
|
377 |
]
|
378 |
gr.Examples(examples=color_examples,
|
@@ -383,10 +385,10 @@ def main():
|
|
383 |
num_segments,
|
384 |
segment_threshold,
|
385 |
inject_interval,
|
|
|
386 |
seed,
|
387 |
color_guidance_weight,
|
388 |
rich_text_input,
|
389 |
-
background_aug,
|
390 |
],
|
391 |
outputs=[
|
392 |
plaintext_result,
|
@@ -395,7 +397,7 @@ def main():
|
|
395 |
token_map,
|
396 |
],
|
397 |
fn=generate,
|
398 |
-
|
399 |
examples_per_page=20)
|
400 |
|
401 |
with gr.Row():
|
@@ -403,13 +405,13 @@ def main():
|
|
403 |
[
|
404 |
'{"ops":[{"insert":"a "},{"attributes":{"font":"mirza"},"insert":"beautiful garden"},{"insert":" with a "},{"attributes":{"font":"roboto"},"insert":"snow mountain in the background"},{"insert":""}]}',
|
405 |
'',
|
406 |
-
|
407 |
-
0.
|
|
|
408 |
0.2,
|
409 |
3,
|
410 |
-
0
|
411 |
None,
|
412 |
-
False
|
413 |
],
|
414 |
[
|
415 |
'{"ops":[{"attributes":{"link":"the awe-inspiring sky and ocean in the style of J.M.W. Turner"},"insert":"the awe-inspiring sky and sea"},{"insert":" by "},{"attributes":{"font":"mirza"},"insert":"a coast with flowers and grasses in spring"}]}',
|
@@ -417,21 +419,21 @@ def main():
|
|
417 |
5,
|
418 |
0.3,
|
419 |
0,
|
|
|
420 |
9,
|
421 |
0.5,
|
422 |
None,
|
423 |
-
False
|
424 |
],
|
425 |
[
|
426 |
'{"ops":[{"insert":"a "},{"attributes":{"font":"slabo"},"insert":"night sky filled with stars"},{"insert":" above a "},{"attributes":{"font":"roboto"},"insert":"turbulent sea with giant waves"}]}',
|
427 |
'',
|
428 |
2,
|
429 |
-
0.
|
|
|
430 |
0,
|
431 |
6,
|
432 |
0.5,
|
433 |
None,
|
434 |
-
False
|
435 |
],
|
436 |
]
|
437 |
gr.Examples(examples=style_examples,
|
@@ -442,10 +444,10 @@ def main():
|
|
442 |
num_segments,
|
443 |
segment_threshold,
|
444 |
inject_interval,
|
|
|
445 |
seed,
|
446 |
color_guidance_weight,
|
447 |
rich_text_input,
|
448 |
-
background_aug,
|
449 |
],
|
450 |
outputs=[
|
451 |
plaintext_result,
|
@@ -454,7 +456,7 @@ def main():
|
|
454 |
token_map,
|
455 |
],
|
456 |
fn=generate,
|
457 |
-
|
458 |
examples_per_page=20)
|
459 |
|
460 |
with gr.Row():
|
@@ -465,10 +467,10 @@ def main():
|
|
465 |
5,
|
466 |
0.3,
|
467 |
0,
|
|
|
468 |
13,
|
469 |
1,
|
470 |
None,
|
471 |
-
False
|
472 |
],
|
473 |
[
|
474 |
'{"ops": [{"insert": "A pizza with pineapple, "}, {"attributes": {"size": "20px"}, "insert": "pepperoni"}, {"insert": ", and mushroom on the top, 4k, photorealistic"}]}',
|
@@ -476,10 +478,10 @@ def main():
|
|
476 |
5,
|
477 |
0.3,
|
478 |
0,
|
|
|
479 |
13,
|
480 |
1,
|
481 |
None,
|
482 |
-
False
|
483 |
],
|
484 |
[
|
485 |
'{"ops": [{"insert": "A pizza with pineapple, pepperoni, and "}, {"attributes": {"size": "70px"}, "insert": "mushroom"}, {"insert": " on the top, 4k, photorealistic"}]}',
|
@@ -487,10 +489,10 @@ def main():
|
|
487 |
5,
|
488 |
0.3,
|
489 |
0,
|
|
|
490 |
13,
|
491 |
1,
|
492 |
None,
|
493 |
-
False
|
494 |
],
|
495 |
]
|
496 |
gr.Examples(examples=size_examples,
|
@@ -501,10 +503,10 @@ def main():
|
|
501 |
num_segments,
|
502 |
segment_threshold,
|
503 |
inject_interval,
|
|
|
504 |
seed,
|
505 |
color_guidance_weight,
|
506 |
rich_text_input,
|
507 |
-
background_aug,
|
508 |
],
|
509 |
outputs=[
|
510 |
plaintext_result,
|
@@ -513,24 +515,24 @@ def main():
|
|
513 |
token_map,
|
514 |
],
|
515 |
fn=generate,
|
516 |
-
|
517 |
examples_per_page=20)
|
518 |
generate_button.click(fn=lambda: gr.update(visible=False), inputs=None, outputs=share_row, queue=False).then(
|
519 |
fn=generate,
|
520 |
inputs=[
|
521 |
text_input,
|
522 |
negative_prompt,
|
523 |
-
height,
|
524 |
-
width,
|
525 |
-
seed,
|
526 |
-
steps,
|
527 |
num_segments,
|
528 |
segment_threshold,
|
529 |
inject_interval,
|
530 |
-
|
|
|
531 |
color_guidance_weight,
|
532 |
rich_text_input,
|
533 |
-
|
|
|
|
|
|
|
534 |
],
|
535 |
outputs=[plaintext_result, richtext_result, segments, token_map],
|
536 |
_js=get_js_data
|
|
|
29 |
|
30 |
canvas_html = """<iframe id='rich-text-root' style='width:100%' height='360px' src='file=rich-text-to-json-iframe.html' frameborder='0' scrolling='no'></iframe>"""
|
31 |
get_js_data = """
|
32 |
+
async (text_input, negative_prompt, num_segments, segment_threshold, inject_interval, inject_background, seed, color_guidance_weight, rich_text_input, height, width, steps, guidance_weights) => {
|
33 |
const richEl = document.getElementById("rich-text-root");
|
34 |
const data = richEl? richEl.contentDocument.body._data : {};
|
35 |
+
return [text_input, negative_prompt, num_segments, segment_threshold, inject_interval, inject_background, seed, color_guidance_weight, JSON.stringify(data), height, width, steps, guidance_weights];
|
36 |
}
|
37 |
"""
|
38 |
set_js_data = """
|
|
|
66 |
def generate(
|
67 |
text_input: str,
|
68 |
negative_text: str,
|
|
|
|
|
|
|
|
|
69 |
num_segments: int,
|
70 |
segment_threshold: float,
|
71 |
inject_interval: float,
|
72 |
+
inject_background: float,
|
73 |
+
seed: int,
|
74 |
color_guidance_weight: float,
|
75 |
rich_text_input: str,
|
76 |
+
height: int,
|
77 |
+
width: int,
|
78 |
+
steps: int,
|
79 |
+
guidance_weight: float,
|
80 |
):
|
81 |
run_dir = 'results/'
|
82 |
os.makedirs(run_dir, exist_ok=True)
|
83 |
# Load region diffusion model.
|
84 |
+
height = int(height) if height else 512
|
85 |
+
width = int(width) if width else 512
|
86 |
steps = 41 if not steps else steps
|
87 |
guidance_weight = 8.5 if not guidance_weight else guidance_weight
|
88 |
+
text_input = rich_text_input if rich_text_input != '' and rich_text_input != None else text_input
|
89 |
+
print('text_input', text_input, width, height, steps, guidance_weight, num_segments, segment_threshold, inject_interval, inject_background, color_guidance_weight, negative_text)
|
90 |
if (text_input == '' or rich_text_input == ''):
|
91 |
raise gr.Error("Please enter some text.")
|
92 |
# parse json to span attributes
|
|
|
132 |
512//8, 512//8, region_target_token_ids[:-1], seed,
|
133 |
base_tokens, segment_threshold=segment_threshold, num_segments=num_segments,
|
134 |
return_vis=True)
|
135 |
+
color_obj_atten_all = torch.zeros_like(color_obj_masks[-1])
|
136 |
+
for obj_mask in color_obj_masks[:-1]:
|
137 |
+
color_obj_atten_all += obj_mask
|
138 |
color_obj_masks = [transforms.functional.resize(color_obj_mask, (height, width),
|
139 |
interpolation=transforms.InterpolationMode.BICUBIC,
|
140 |
antialias=True)
|
141 |
for color_obj_mask in color_obj_masks]
|
142 |
text_format_dict['color_obj_atten'] = color_obj_masks
|
143 |
+
text_format_dict['color_obj_atten_all'] = color_obj_atten_all
|
144 |
model.remove_tokenmap_hooks()
|
145 |
|
146 |
# generate image from rich text
|
147 |
begin_time = time.time()
|
148 |
seed_everything(seed)
|
|
|
|
|
|
|
|
|
149 |
rich_img = model.prompt_to_img(region_text_prompts, [negative_text],
|
150 |
height=height, width=width, num_inference_steps=steps,
|
151 |
guidance_scale=guidance_weight, use_guidance=use_grad_guidance,
|
152 |
text_format_dict=text_format_dict, inject_selfattn=inject_interval,
|
153 |
+
inject_background=inject_background)
|
154 |
print('time lapses to generate image from rich text: %.4f' %
|
155 |
(time.time()-begin_time))
|
156 |
return [plain_img[0], rich_img[0], segments_vis, token_maps]
|
|
|
191 |
maximum=1,
|
192 |
step=0.01,
|
193 |
value=0.)
|
194 |
+
inject_background = gr.Slider(label='Unformatted token preservation',
|
195 |
+
info='(To affect less the tokens without any rich-text attributes, increase this.)',
|
196 |
+
minimum=0,
|
197 |
+
maximum=1,
|
198 |
+
step=0.01,
|
199 |
+
value=0.3)
|
200 |
color_guidance_weight = gr.Slider(label='Color weight',
|
201 |
info='(To obtain more precise color, increase this, while too large value may cause artifacts.)',
|
202 |
minimum=0,
|
|
|
215 |
value=6,
|
216 |
elem_id="seed"
|
217 |
)
|
|
|
|
|
|
|
|
|
218 |
with gr.Accordion('Other Parameters', open=False):
|
219 |
steps = gr.Slider(label='Number of Steps',
|
220 |
minimum=0,
|
|
|
268 |
5,
|
269 |
0.3,
|
270 |
0,
|
271 |
+
0.5,
|
272 |
6,
|
273 |
+
0,
|
274 |
None,
|
|
|
275 |
],
|
276 |
[
|
277 |
+
'{"ops":[{"insert":"A "},{"attributes":{"link":"Thor Kitchen 30 Inch Wide Freestanding Gas Range with Automatic Re-Ignition System"},"insert":"kitchen island"},{"insert":" next to a "},{"attributes":{"link":"an open refrigerator stocked with fresh produce, dairy products, and beverages. "},"insert":"refrigerator"},{"insert":", by James McDonald and Joarc Architects, home, interior, octane render, deviantart, cinematic, key art, hyperrealism, sun light, sunrays, canon eos c 300, ƒ 1.8, 35 mm, 8k, medium - format print"}]}',
|
278 |
'',
|
279 |
+
7,
|
280 |
0.5,
|
281 |
0,
|
282 |
+
0.5,
|
283 |
6,
|
284 |
+
0,
|
285 |
None,
|
|
|
286 |
],
|
287 |
[
|
288 |
'{"ops":[{"insert":"A "},{"attributes":{"link":"Happy Kung fu panda art, elder, asian art, volumetric lighting, dramatic scene, ultra detailed, realism, chinese"},"insert":"panda"},{"insert":" standing on a cliff by a waterfall, wildlife photography, photograph, high quality, wildlife, f 1.8, soft focus, 8k, national geographic, award - winning photograph by nick nichols"}]}',
|
289 |
'',
|
290 |
+
5,
|
291 |
0.3,
|
292 |
0,
|
293 |
+
0.1,
|
294 |
4,
|
295 |
+
0,
|
296 |
None,
|
|
|
297 |
],
|
298 |
]
|
299 |
|
|
|
305 |
num_segments,
|
306 |
segment_threshold,
|
307 |
inject_interval,
|
308 |
+
inject_background,
|
309 |
seed,
|
310 |
color_guidance_weight,
|
311 |
rich_text_input,
|
|
|
312 |
],
|
313 |
outputs=[
|
314 |
plaintext_result,
|
|
|
317 |
token_map,
|
318 |
],
|
319 |
fn=generate,
|
320 |
+
cache_examples=True,
|
321 |
examples_per_page=20)
|
322 |
with gr.Row():
|
323 |
color_examples = [
|
324 |
[
|
325 |
+
'{"ops":[{"insert":"a beautifule girl with big eye, skin, and long "},{"attributes":{"color":"#04a704"},"insert":"hair"},{"insert":", t-shirt, bursting with vivid color, intricate, elegant, highly detailed, photorealistic, digital painting, artstation, illustration, concept art."}]}',
|
326 |
'lowres, had anatomy, bad hands, cropped, worst quality',
|
327 |
+
11,
|
328 |
+
0.3,
|
329 |
+
0.3,
|
330 |
0.3,
|
331 |
6,
|
332 |
0.5,
|
333 |
None,
|
|
|
334 |
],
|
335 |
[
|
336 |
+
'{"ops":[{"insert":"a beautifule girl with big eye, skin, and long "},{"attributes":{"color":"#999999"},"insert":"hair"},{"insert":", t-shirt, bursting with vivid color, intricate, elegant, highly detailed, photorealistic, digital painting, artstation, illustration, concept art."}]}',
|
337 |
'lowres, had anatomy, bad hands, cropped, worst quality',
|
338 |
+
11,
|
339 |
+
0.3,
|
340 |
+
0.3,
|
341 |
0.3,
|
342 |
6,
|
343 |
+
0.5,
|
344 |
None,
|
|
|
345 |
],
|
346 |
[
|
347 |
'{"ops":[{"insert":"a Gothic "},{"attributes":{"color":"#FD6C9E"},"insert":"church"},{"insert":" in a the sunset with a beautiful landscape in the background."}]}',
|
348 |
'',
|
349 |
+
10,
|
350 |
+
0.4,
|
351 |
0.5,
|
352 |
+
0.3,
|
353 |
6,
|
354 |
0.5,
|
355 |
None,
|
|
|
356 |
],
|
357 |
[
|
358 |
'{"ops":[{"insert":"A mesmerizing sight that captures the beauty of a "},{"attributes":{"color":"#4775fc"},"insert":"rose"},{"insert":" blooming, close up"}]}',
|
|
|
360 |
3,
|
361 |
0.3,
|
362 |
0,
|
363 |
+
0,
|
364 |
9,
|
365 |
1,
|
366 |
None,
|
|
|
367 |
],
|
368 |
[
|
369 |
'{"ops":[{"insert":"A "},{"attributes":{"color":"#FFD700"},"insert":"marble statue of a wolf\'s head and shoulder"},{"insert":", surrounded by colorful flowers michelangelo, detailed, intricate, full of color, led lighting, trending on artstation, 4 k, hyperrealistic, 3 5 mm, focused, extreme details, unreal engine 5, masterpiece "}]}',
|
370 |
'',
|
371 |
5,
|
372 |
+
0.4,
|
373 |
+
0.3,
|
374 |
0.3,
|
|
|
375 |
5,
|
376 |
0.6,
|
377 |
None,
|
|
|
378 |
],
|
379 |
]
|
380 |
gr.Examples(examples=color_examples,
|
|
|
385 |
num_segments,
|
386 |
segment_threshold,
|
387 |
inject_interval,
|
388 |
+
inject_background,
|
389 |
seed,
|
390 |
color_guidance_weight,
|
391 |
rich_text_input,
|
|
|
392 |
],
|
393 |
outputs=[
|
394 |
plaintext_result,
|
|
|
397 |
token_map,
|
398 |
],
|
399 |
fn=generate,
|
400 |
+
cache_examples=True,
|
401 |
examples_per_page=20)
|
402 |
|
403 |
with gr.Row():
|
|
|
405 |
[
|
406 |
'{"ops":[{"insert":"a "},{"attributes":{"font":"mirza"},"insert":"beautiful garden"},{"insert":" with a "},{"attributes":{"font":"roboto"},"insert":"snow mountain in the background"},{"insert":""}]}',
|
407 |
'',
|
408 |
+
10,
|
409 |
+
0.4,
|
410 |
+
0,
|
411 |
0.2,
|
412 |
3,
|
413 |
+
0,
|
414 |
None,
|
|
|
415 |
],
|
416 |
[
|
417 |
'{"ops":[{"attributes":{"link":"the awe-inspiring sky and ocean in the style of J.M.W. Turner"},"insert":"the awe-inspiring sky and sea"},{"insert":" by "},{"attributes":{"font":"mirza"},"insert":"a coast with flowers and grasses in spring"}]}',
|
|
|
419 |
5,
|
420 |
0.3,
|
421 |
0,
|
422 |
+
0,
|
423 |
9,
|
424 |
0.5,
|
425 |
None,
|
|
|
426 |
],
|
427 |
[
|
428 |
'{"ops":[{"insert":"a "},{"attributes":{"font":"slabo"},"insert":"night sky filled with stars"},{"insert":" above a "},{"attributes":{"font":"roboto"},"insert":"turbulent sea with giant waves"}]}',
|
429 |
'',
|
430 |
2,
|
431 |
+
0.35,
|
432 |
+
0,
|
433 |
0,
|
434 |
6,
|
435 |
0.5,
|
436 |
None,
|
|
|
437 |
],
|
438 |
]
|
439 |
gr.Examples(examples=style_examples,
|
|
|
444 |
num_segments,
|
445 |
segment_threshold,
|
446 |
inject_interval,
|
447 |
+
inject_background,
|
448 |
seed,
|
449 |
color_guidance_weight,
|
450 |
rich_text_input,
|
|
|
451 |
],
|
452 |
outputs=[
|
453 |
plaintext_result,
|
|
|
456 |
token_map,
|
457 |
],
|
458 |
fn=generate,
|
459 |
+
cache_examples=True,
|
460 |
examples_per_page=20)
|
461 |
|
462 |
with gr.Row():
|
|
|
467 |
5,
|
468 |
0.3,
|
469 |
0,
|
470 |
+
0,
|
471 |
13,
|
472 |
1,
|
473 |
None,
|
|
|
474 |
],
|
475 |
[
|
476 |
'{"ops": [{"insert": "A pizza with pineapple, "}, {"attributes": {"size": "20px"}, "insert": "pepperoni"}, {"insert": ", and mushroom on the top, 4k, photorealistic"}]}',
|
|
|
478 |
5,
|
479 |
0.3,
|
480 |
0,
|
481 |
+
0,
|
482 |
13,
|
483 |
1,
|
484 |
None,
|
|
|
485 |
],
|
486 |
[
|
487 |
'{"ops": [{"insert": "A pizza with pineapple, pepperoni, and "}, {"attributes": {"size": "70px"}, "insert": "mushroom"}, {"insert": " on the top, 4k, photorealistic"}]}',
|
|
|
489 |
5,
|
490 |
0.3,
|
491 |
0,
|
492 |
+
0,
|
493 |
13,
|
494 |
1,
|
495 |
None,
|
|
|
496 |
],
|
497 |
]
|
498 |
gr.Examples(examples=size_examples,
|
|
|
503 |
num_segments,
|
504 |
segment_threshold,
|
505 |
inject_interval,
|
506 |
+
inject_background,
|
507 |
seed,
|
508 |
color_guidance_weight,
|
509 |
rich_text_input,
|
|
|
510 |
],
|
511 |
outputs=[
|
512 |
plaintext_result,
|
|
|
515 |
token_map,
|
516 |
],
|
517 |
fn=generate,
|
518 |
+
cache_examples=True,
|
519 |
examples_per_page=20)
|
520 |
generate_button.click(fn=lambda: gr.update(visible=False), inputs=None, outputs=share_row, queue=False).then(
|
521 |
fn=generate,
|
522 |
inputs=[
|
523 |
text_input,
|
524 |
negative_prompt,
|
|
|
|
|
|
|
|
|
525 |
num_segments,
|
526 |
segment_threshold,
|
527 |
inject_interval,
|
528 |
+
inject_background,
|
529 |
+
seed,
|
530 |
color_guidance_weight,
|
531 |
rich_text_input,
|
532 |
+
height,
|
533 |
+
width,
|
534 |
+
steps,
|
535 |
+
guidance_weight,
|
536 |
],
|
537 |
outputs=[plaintext_result, richtext_result, segments, token_map],
|
538 |
_js=get_js_data
|
models/region_diffusion.py
CHANGED
@@ -84,13 +84,13 @@ class RegionDiffusion(nn.Module):
|
|
84 |
return text_embeddings
|
85 |
|
86 |
def produce_latents(self, text_embeddings, height=512, width=512, num_inference_steps=50, guidance_scale=7.5,
|
87 |
-
latents=None, use_guidance=False, text_format_dict={}, inject_selfattn=0,
|
88 |
|
89 |
if latents is None:
|
90 |
latents = torch.randn(
|
91 |
(1, self.unet.in_channels, height // 8, width // 8), device=self.device)
|
92 |
|
93 |
-
if inject_selfattn > 0:
|
94 |
latents_reference = latents.clone().detach()
|
95 |
self.scheduler.set_timesteps(num_inference_steps)
|
96 |
n_styles = text_embeddings.shape[0]-1
|
@@ -102,11 +102,12 @@ class RegionDiffusion(nn.Module):
|
|
102 |
with torch.no_grad():
|
103 |
# tokens without any attributes
|
104 |
feat_inject_step = t > (1-inject_selfattn) * 1000
|
|
|
105 |
noise_pred_uncond_cur = self.unet(latents, t, encoder_hidden_states=text_embeddings[:1],
|
106 |
-
|
107 |
noise_pred_text_cur = self.unet(latents, t, encoder_hidden_states=text_embeddings[-1:],
|
108 |
text_format_dict=text_format_dict)['sample']
|
109 |
-
if inject_selfattn > 0:
|
110 |
noise_pred_uncond_refer = self.unet(latents_reference, t, encoder_hidden_states=text_embeddings[:1],
|
111 |
text_format_dict={})['sample']
|
112 |
self.register_selfattn_hooks(feat_inject_step)
|
@@ -117,33 +118,18 @@ class RegionDiffusion(nn.Module):
|
|
117 |
noise_pred_text = noise_pred_text_cur * self.masks[-1]
|
118 |
# tokens with attributes
|
119 |
for style_i, mask in enumerate(self.masks[:-1]):
|
120 |
-
if t > bg_aug_end:
|
121 |
-
rand_rgb = torch.rand([1, 3, 1, 1]).cuda()
|
122 |
-
black_background = torch.ones(
|
123 |
-
[1, 3, height, width]).cuda()*rand_rgb
|
124 |
-
black_latent = self.encode_imgs(
|
125 |
-
black_background)
|
126 |
-
noise = torch.randn_like(black_latent)
|
127 |
-
black_latent_noisy = self.scheduler.add_noise(
|
128 |
-
black_latent, noise, t)
|
129 |
-
masked_latent = (
|
130 |
-
mask > 0.001) * latents + (mask < 0.001) * black_latent_noisy
|
131 |
-
noise_pred_uncond_cur = self.unet(masked_latent, t, encoder_hidden_states=text_embeddings[:1],
|
132 |
-
text_format_dict={})['sample']
|
133 |
-
else:
|
134 |
-
masked_latent = latents
|
135 |
self.register_replacement_hooks(feat_inject_step)
|
136 |
-
noise_pred_text_cur = self.unet(
|
137 |
text_format_dict={})['sample']
|
138 |
self.remove_replacement_hooks()
|
139 |
noise_pred_uncond = noise_pred_uncond + noise_pred_uncond_cur*mask
|
140 |
noise_pred_text = noise_pred_text + noise_pred_text_cur*mask
|
141 |
-
|
142 |
# perform classifier-free guidance
|
143 |
noise_pred = noise_pred_uncond + guidance_scale * \
|
144 |
(noise_pred_text - noise_pred_uncond)
|
145 |
|
146 |
-
if inject_selfattn > 0:
|
147 |
noise_pred_refer = noise_pred_uncond_refer + guidance_scale * \
|
148 |
(noise_pred_text_refer - noise_pred_uncond_refer)
|
149 |
|
@@ -174,12 +160,15 @@ class RegionDiffusion(nn.Module):
|
|
174 |
imgs*attn_map[:, 0]).sum(2).sum(2)/attn_map[:, 0].sum()
|
175 |
loss = self.color_loss(
|
176 |
avg_rgb, rgb_val[:, :, 0, 0])*100
|
177 |
-
# print(loss)
|
178 |
loss_total += loss
|
179 |
loss_total.backward()
|
180 |
latents = (
|
181 |
-
latents - latents.grad * text_format_dict['color_guidance_weight'] *
|
182 |
|
|
|
|
|
|
|
|
|
183 |
return latents
|
184 |
|
185 |
def predict_x0(self, x_t, eps_t, t):
|
@@ -255,7 +244,7 @@ class RegionDiffusion(nn.Module):
|
|
255 |
return latents
|
256 |
|
257 |
def prompt_to_img(self, prompts, negative_prompts='', height=512, width=512, num_inference_steps=50,
|
258 |
-
guidance_scale=7.5, latents=None, text_format_dict={}, use_guidance=False, inject_selfattn=0,
|
259 |
|
260 |
if isinstance(prompts, str):
|
261 |
prompts = [prompts]
|
@@ -271,7 +260,7 @@ class RegionDiffusion(nn.Module):
|
|
271 |
latents = self.produce_latents(text_embeds, height=height, width=width, latents=latents,
|
272 |
num_inference_steps=num_inference_steps, guidance_scale=guidance_scale,
|
273 |
use_guidance=use_guidance, text_format_dict=text_format_dict,
|
274 |
-
inject_selfattn=inject_selfattn,
|
275 |
# Img latents -> imgs
|
276 |
imgs = self.decode_latents(latents) # [1, 3, 512, 512]
|
277 |
|
@@ -345,8 +334,6 @@ class RegionDiffusion(nn.Module):
|
|
345 |
"""
|
346 |
# out[0] - final output of residual layer
|
347 |
# out[1] - residual hidden feature
|
348 |
-
# import ipdb
|
349 |
-
# ipdb.set_trace()
|
350 |
assert out[1].shape[-1] == 16
|
351 |
activations[name] = out[1].detach()
|
352 |
attention_dict = collections.defaultdict(list)
|
|
|
84 |
return text_embeddings
|
85 |
|
86 |
def produce_latents(self, text_embeddings, height=512, width=512, num_inference_steps=50, guidance_scale=7.5,
|
87 |
+
latents=None, use_guidance=False, text_format_dict={}, inject_selfattn=0, inject_background=0):
|
88 |
|
89 |
if latents is None:
|
90 |
latents = torch.randn(
|
91 |
(1, self.unet.in_channels, height // 8, width // 8), device=self.device)
|
92 |
|
93 |
+
if inject_selfattn > 0 or inject_background > 0:
|
94 |
latents_reference = latents.clone().detach()
|
95 |
self.scheduler.set_timesteps(num_inference_steps)
|
96 |
n_styles = text_embeddings.shape[0]-1
|
|
|
102 |
with torch.no_grad():
|
103 |
# tokens without any attributes
|
104 |
feat_inject_step = t > (1-inject_selfattn) * 1000
|
105 |
+
background_inject_step = i == int(inject_background * len(self.scheduler.timesteps)) and inject_background > 0
|
106 |
noise_pred_uncond_cur = self.unet(latents, t, encoder_hidden_states=text_embeddings[:1],
|
107 |
+
text_format_dict={})['sample']
|
108 |
noise_pred_text_cur = self.unet(latents, t, encoder_hidden_states=text_embeddings[-1:],
|
109 |
text_format_dict=text_format_dict)['sample']
|
110 |
+
if inject_selfattn > 0 or inject_background > 0:
|
111 |
noise_pred_uncond_refer = self.unet(latents_reference, t, encoder_hidden_states=text_embeddings[:1],
|
112 |
text_format_dict={})['sample']
|
113 |
self.register_selfattn_hooks(feat_inject_step)
|
|
|
118 |
noise_pred_text = noise_pred_text_cur * self.masks[-1]
|
119 |
# tokens with attributes
|
120 |
for style_i, mask in enumerate(self.masks[:-1]):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
121 |
self.register_replacement_hooks(feat_inject_step)
|
122 |
+
noise_pred_text_cur = self.unet(latents, t, encoder_hidden_states=text_embeddings[style_i+1:style_i+2],
|
123 |
text_format_dict={})['sample']
|
124 |
self.remove_replacement_hooks()
|
125 |
noise_pred_uncond = noise_pred_uncond + noise_pred_uncond_cur*mask
|
126 |
noise_pred_text = noise_pred_text + noise_pred_text_cur*mask
|
127 |
+
|
128 |
# perform classifier-free guidance
|
129 |
noise_pred = noise_pred_uncond + guidance_scale * \
|
130 |
(noise_pred_text - noise_pred_uncond)
|
131 |
|
132 |
+
if inject_selfattn > 0 or inject_background > 0:
|
133 |
noise_pred_refer = noise_pred_uncond_refer + guidance_scale * \
|
134 |
(noise_pred_text_refer - noise_pred_uncond_refer)
|
135 |
|
|
|
160 |
imgs*attn_map[:, 0]).sum(2).sum(2)/attn_map[:, 0].sum()
|
161 |
loss = self.color_loss(
|
162 |
avg_rgb, rgb_val[:, :, 0, 0])*100
|
|
|
163 |
loss_total += loss
|
164 |
loss_total.backward()
|
165 |
latents = (
|
166 |
+
latents - latents.grad * text_format_dict['color_guidance_weight'] * text_format_dict['color_obj_atten_all']).detach().clone()
|
167 |
|
168 |
+
# apply background injection
|
169 |
+
if background_inject_step:
|
170 |
+
latents = latents_reference * self.masks[-1] + latents * \
|
171 |
+
(1-self.masks[-1])
|
172 |
return latents
|
173 |
|
174 |
def predict_x0(self, x_t, eps_t, t):
|
|
|
244 |
return latents
|
245 |
|
246 |
def prompt_to_img(self, prompts, negative_prompts='', height=512, width=512, num_inference_steps=50,
|
247 |
+
guidance_scale=7.5, latents=None, text_format_dict={}, use_guidance=False, inject_selfattn=0, inject_background=0):
|
248 |
|
249 |
if isinstance(prompts, str):
|
250 |
prompts = [prompts]
|
|
|
260 |
latents = self.produce_latents(text_embeds, height=height, width=width, latents=latents,
|
261 |
num_inference_steps=num_inference_steps, guidance_scale=guidance_scale,
|
262 |
use_guidance=use_guidance, text_format_dict=text_format_dict,
|
263 |
+
inject_selfattn=inject_selfattn, inject_background=inject_background) # [1, 4, 64, 64]
|
264 |
# Img latents -> imgs
|
265 |
imgs = self.decode_latents(latents) # [1, 3, 512, 512]
|
266 |
|
|
|
334 |
"""
|
335 |
# out[0] - final output of residual layer
|
336 |
# out[1] - residual hidden feature
|
|
|
|
|
337 |
assert out[1].shape[-1] == 16
|
338 |
activations[name] = out[1].detach()
|
339 |
attention_dict = collections.defaultdict(list)
|
utils/attention_utils.py
CHANGED
@@ -6,25 +6,26 @@ import seaborn as sns
|
|
6 |
import torch
|
7 |
import torchvision
|
8 |
|
9 |
-
from
|
|
|
10 |
|
11 |
SelfAttentionLayers = [
|
12 |
-
|
13 |
-
|
14 |
'down_blocks.1.attentions.0.transformer_blocks.0.attn1',
|
15 |
-
|
16 |
'down_blocks.2.attentions.0.transformer_blocks.0.attn1',
|
17 |
'down_blocks.2.attentions.1.transformer_blocks.0.attn1',
|
18 |
'mid_block.attentions.0.transformer_blocks.0.attn1',
|
19 |
'up_blocks.1.attentions.0.transformer_blocks.0.attn1',
|
20 |
'up_blocks.1.attentions.1.transformer_blocks.0.attn1',
|
21 |
'up_blocks.1.attentions.2.transformer_blocks.0.attn1',
|
22 |
-
|
23 |
'up_blocks.2.attentions.1.transformer_blocks.0.attn1',
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
]
|
29 |
|
30 |
|
@@ -208,8 +209,8 @@ def get_token_maps_deprecated(attention_maps, save_dir, width, height, obj_token
|
|
208 |
return attention_maps_averaged_normalized, token_maps_vis
|
209 |
|
210 |
|
211 |
-
def get_token_maps(selfattn_maps, crossattn_maps, n_maps, save_dir, width, height, obj_tokens,
|
212 |
-
preprocess=False, segment_threshold=0.
|
213 |
r"""Function to visualize attention maps.
|
214 |
Args:
|
215 |
save_dir (str): Path to save attention maps
|
@@ -219,9 +220,11 @@ def get_token_maps(selfattn_maps, crossattn_maps, n_maps, save_dir, width, heigh
|
|
219 |
|
220 |
# create the segmentation mask using self-attention maps
|
221 |
resolution = 32
|
222 |
-
attn_maps_1024 = {8: [], 16: [], 32: []}
|
223 |
for attn_map in selfattn_maps.values():
|
224 |
resolution_map = np.sqrt(attn_map.shape[1]).astype(int)
|
|
|
|
|
225 |
attn_map = attn_map.reshape(
|
226 |
1, resolution_map, resolution_map, resolution_map**2).permute([3, 0, 1, 2])
|
227 |
attn_map = torch.nn.functional.interpolate(attn_map, (resolution, resolution),
|
@@ -229,10 +232,15 @@ def get_token_maps(selfattn_maps, crossattn_maps, n_maps, save_dir, width, heigh
|
|
229 |
attn_maps_1024[resolution_map].append(attn_map.permute([1, 2, 3, 0]).reshape(
|
230 |
1, resolution**2, resolution_map**2))
|
231 |
attn_maps_1024 = torch.cat([torch.cat(v).mean(0).cpu()
|
232 |
-
for v in attn_maps_1024.values()], -1).numpy()
|
233 |
-
|
234 |
-
|
235 |
-
|
|
|
|
|
|
|
|
|
|
|
236 |
clusters = clusters.reshape(resolution, resolution)
|
237 |
fig = plt.figure()
|
238 |
plt.imshow(clusters)
|
@@ -258,6 +266,10 @@ def get_token_maps(selfattn_maps, crossattn_maps, n_maps, save_dir, width, heigh
|
|
258 |
|
259 |
cross_attn_maps_1024 = torch.cat(
|
260 |
cross_attn_maps_1024).mean(0).cpu().numpy()
|
|
|
|
|
|
|
|
|
261 |
normalized_span_maps = []
|
262 |
for token_ids in obj_tokens:
|
263 |
span_token_maps = cross_attn_maps_1024[:, :, token_ids.numpy()]
|
@@ -297,7 +309,7 @@ def get_token_maps(selfattn_maps, crossattn_maps, n_maps, save_dir, width, heigh
|
|
297 |
foreground_token_maps = [token_map[None, :, :]
|
298 |
for token_map in foreground_token_maps]
|
299 |
token_maps_vis = plot_attention_maps([foreground_token_maps, resized_token_maps], obj_tokens,
|
300 |
-
save_dir,
|
301 |
resized_token_maps = [token_map.unsqueeze(1).repeat(
|
302 |
[1, 4, 1, 1]).to(attn_map.dtype).cuda() for token_map in resized_token_maps]
|
303 |
if return_vis:
|
|
|
6 |
import torch
|
7 |
import torchvision
|
8 |
|
9 |
+
from utils.richtext_utils import seed_everything
|
10 |
+
from sklearn.cluster import SpectralClustering
|
11 |
|
12 |
SelfAttentionLayers = [
|
13 |
+
'down_blocks.0.attentions.0.transformer_blocks.0.attn1',
|
14 |
+
'down_blocks.0.attentions.1.transformer_blocks.0.attn1',
|
15 |
'down_blocks.1.attentions.0.transformer_blocks.0.attn1',
|
16 |
+
'down_blocks.1.attentions.1.transformer_blocks.0.attn1',
|
17 |
'down_blocks.2.attentions.0.transformer_blocks.0.attn1',
|
18 |
'down_blocks.2.attentions.1.transformer_blocks.0.attn1',
|
19 |
'mid_block.attentions.0.transformer_blocks.0.attn1',
|
20 |
'up_blocks.1.attentions.0.transformer_blocks.0.attn1',
|
21 |
'up_blocks.1.attentions.1.transformer_blocks.0.attn1',
|
22 |
'up_blocks.1.attentions.2.transformer_blocks.0.attn1',
|
23 |
+
'up_blocks.2.attentions.0.transformer_blocks.0.attn1',
|
24 |
'up_blocks.2.attentions.1.transformer_blocks.0.attn1',
|
25 |
+
'up_blocks.2.attentions.2.transformer_blocks.0.attn1',
|
26 |
+
'up_blocks.3.attentions.0.transformer_blocks.0.attn1',
|
27 |
+
'up_blocks.3.attentions.1.transformer_blocks.0.attn1',
|
28 |
+
'up_blocks.3.attentions.2.transformer_blocks.0.attn1',
|
29 |
]
|
30 |
|
31 |
|
|
|
209 |
return attention_maps_averaged_normalized, token_maps_vis
|
210 |
|
211 |
|
212 |
+
def get_token_maps(selfattn_maps, crossattn_maps, n_maps, save_dir, width, height, obj_tokens, seed=0, tokens_vis=None,
|
213 |
+
preprocess=False, segment_threshold=0.3, num_segments=5, return_vis=False, save_attn=False):
|
214 |
r"""Function to visualize attention maps.
|
215 |
Args:
|
216 |
save_dir (str): Path to save attention maps
|
|
|
220 |
|
221 |
# create the segmentation mask using self-attention maps
|
222 |
resolution = 32
|
223 |
+
attn_maps_1024 = {8: [], 16: [], 32: [], 64: []}
|
224 |
for attn_map in selfattn_maps.values():
|
225 |
resolution_map = np.sqrt(attn_map.shape[1]).astype(int)
|
226 |
+
if resolution_map != resolution:
|
227 |
+
continue
|
228 |
attn_map = attn_map.reshape(
|
229 |
1, resolution_map, resolution_map, resolution_map**2).permute([3, 0, 1, 2])
|
230 |
attn_map = torch.nn.functional.interpolate(attn_map, (resolution, resolution),
|
|
|
232 |
attn_maps_1024[resolution_map].append(attn_map.permute([1, 2, 3, 0]).reshape(
|
233 |
1, resolution**2, resolution_map**2))
|
234 |
attn_maps_1024 = torch.cat([torch.cat(v).mean(0).cpu()
|
235 |
+
for v in attn_maps_1024.values() if len(v) > 0], -1).numpy()
|
236 |
+
if save_attn:
|
237 |
+
print('saving self-attention maps...', attn_maps_1024.shape)
|
238 |
+
torch.save(torch.from_numpy(attn_maps_1024),
|
239 |
+
'results/maps/selfattn_maps.pth')
|
240 |
+
seed_everything(seed)
|
241 |
+
sc = SpectralClustering(num_segments, affinity='precomputed', n_init=100,
|
242 |
+
assign_labels='kmeans')
|
243 |
+
clusters = sc.fit_predict(attn_maps_1024)
|
244 |
clusters = clusters.reshape(resolution, resolution)
|
245 |
fig = plt.figure()
|
246 |
plt.imshow(clusters)
|
|
|
266 |
|
267 |
cross_attn_maps_1024 = torch.cat(
|
268 |
cross_attn_maps_1024).mean(0).cpu().numpy()
|
269 |
+
if save_attn:
|
270 |
+
print('saving cross-attention maps...', cross_attn_maps_1024.shape)
|
271 |
+
torch.save(torch.from_numpy(cross_attn_maps_1024),
|
272 |
+
'results/maps/crossattn_maps.pth')
|
273 |
normalized_span_maps = []
|
274 |
for token_ids in obj_tokens:
|
275 |
span_token_maps = cross_attn_maps_1024[:, :, token_ids.numpy()]
|
|
|
309 |
foreground_token_maps = [token_map[None, :, :]
|
310 |
for token_map in foreground_token_maps]
|
311 |
token_maps_vis = plot_attention_maps([foreground_token_maps, resized_token_maps], obj_tokens,
|
312 |
+
save_dir, seed, tokens_vis)
|
313 |
resized_token_maps = [token_map.unsqueeze(1).repeat(
|
314 |
[1, 4, 1, 1]).to(attn_map.dtype).cuda() for token_map in resized_token_maps]
|
315 |
if return_vis:
|