File size: 27,530 Bytes
70a1183
de25831
8c87d48
 
de25831
4409449
 
 
 
 
f0a0274
4409449
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8554568
4409449
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f0a0274
 
4409449
f0a0274
4409449
 
f0a0274
 
a361a0e
44c4801
4d2b5fc
f0a0274
4409449
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2da4702
4409449
e804e78
 
4409449
8554568
 
 
4409449
 
 
 
 
 
8554568
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e804e78
 
 
 
 
 
 
 
8554568
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
fdf81a7
8554568
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
fdf81a7
8554568
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
54b9ded
 
a8de91e
54b9ded
8554568
4409449
 
 
 
 
 
 
 
 
 
 
 
9f3bb67
3d5431e
 
 
 
 
 
 
 
 
9f3bb67
4409449
 
 
 
 
e804e78
4409449
 
 
 
 
 
 
 
 
8554568
4409449
 
 
 
 
 
 
 
 
 
8554568
7953440
4409449
 
 
 
 
 
 
 
8554568
4409449
 
 
 
 
 
 
 
 
b625c80
4409449
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0420e25
 
 
 
8554568
e804e78
8554568
 
 
 
 
 
 
 
 
4409449
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
import os
os.system('pip install /home/user/app/pyrender')
os.system('pip install pyglet==1.4.0a1')
os.system('pip install triangle==20220202')

import gradio as gr
import torch
import time
import numpy as np
import pytorch_lightning as pl
import subprocess
from pathlib import Path
from mGPT.data.build_data import build_data
from mGPT.models.build_model import build_model
from mGPT.config import parse_args
from transformers import WhisperProcessor, WhisperForConditionalGeneration
import librosa
from huggingface_hub import snapshot_download

# Load model
cfg = parse_args(phase="webui")  # parse config file
cfg.FOLDER = 'cache'
output_dir = Path(cfg.FOLDER)
output_dir.mkdir(parents=True, exist_ok=True)
pl.seed_everything(cfg.SEED_VALUE)
if torch.cuda.is_available():
    device = torch.device("cuda")
else:
    device = torch.device("cpu")

model_path = snapshot_download(repo_id="bill-jiang/MotionGPT-base")

datamodule = build_data(cfg, phase="test")
model = build_model(cfg, datamodule)
state_dict = torch.load(f'{model_path}/motiongpt_s3_h3d.tar',
                        map_location="cpu")["state_dict"]
model.load_state_dict(state_dict)
model.to(device)

audio_processor = WhisperProcessor.from_pretrained(cfg.model.whisper_path)
audio_model = WhisperForConditionalGeneration.from_pretrained(
    cfg.model.whisper_path).to(device)
forced_decoder_ids_zh = audio_processor.get_decoder_prompt_ids(
    language="zh", task="translate")
forced_decoder_ids_en = audio_processor.get_decoder_prompt_ids(
    language="en", task="translate")

# HTML Style
Video_Components = """
<div class="side-video" style="position: relative;">
    <video width="340" autoplay loop>
        <source src="file/{video_path}" type="video/mp4">
    </video>
    <a class="videodl-button" href="file/{video_path}" download="{video_fname}" title="Download Video">
        <svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="#000000" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-video"><path d="m22 8-6 4 6 4V8Z"/><rect width="14" height="12" x="2" y="6" rx="2" ry="2"/></svg>
    </a>
    <a class="npydl-button" href="file/{motion_path}" download="{motion_fname}" title="Download Motion">
        <svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="#000000" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-file-box"><path d="M14.5 22H18a2 2 0 0 0 2-2V7.5L14.5 2H6a2 2 0 0 0-2 2v4"/><polyline points="14 2 14 8 20 8"/><path d="M2.97 13.12c-.6.36-.97 1.02-.97 1.74v3.28c0 .72.37 1.38.97 1.74l3 1.83c.63.39 1.43.39 2.06 0l3-1.83c.6-.36.97-1.02.97-1.74v-3.28c0-.72-.37-1.38-.97-1.74l-3-1.83a1.97 1.97 0 0 0-2.06 0l-3 1.83Z"/><path d="m7 17-4.74-2.85"/><path d="m7 17 4.74-2.85"/><path d="M7 17v5"/></svg>
    </a>
</div>
"""

Video_Components_example = """
<div class="side-video" style="position: relative;">
    <video width="340" autoplay loop controls>
        <source src="file/{video_path}" type="video/mp4">
    </video>
    <a class="npydl-button" href="file/{video_path}" download="{video_fname}" title="Download Video">
        <svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-video"><path d="m22 8-6 4 6 4V8Z"/><rect width="14" height="12" x="2" y="6" rx="2" ry="2"/></svg>
    </a>
</div>
"""

Text_Components = """
<h3 class="side-content" >{msg}</h3>
"""


def motion_token_to_string(motion_token, lengths, codebook_size=512):
    motion_string = []
    for i in range(motion_token.shape[0]):
        motion_i = motion_token[i].cpu(
        ) if motion_token.device.type == 'cuda' else motion_token[i]
        motion_list = motion_i.tolist()[:lengths[i]]
        motion_string.append(
            (f'<motion_id_{codebook_size}>' +
             ''.join([f'<motion_id_{int(i)}>' for i in motion_list]) +
             f'<motion_id_{codebook_size + 1}>'))
    return motion_string


def render_motion(data, feats, method='fast'):
    fname = time.strftime("%Y-%m-%d-%H_%M_%S", time.localtime(
        time.time())) + str(np.random.randint(10000, 99999))
    video_fname = fname + '.mp4'
    feats_fname = f"{fname}_feats" + '.npy'
    data_fname = f"{fname}_joints" + '.npy'
    output_npy_path = os.path.join(output_dir, feats_fname)
    output_joints_path = os.path.join(output_dir, data_fname)
    output_mp4_path = os.path.join(output_dir, video_fname)
    np.save(output_npy_path, feats)
    np.save(output_joints_path, data)
    
    render_cmd = ["python", "-m", "render", "--joints_path", output_joints_path, "--method", method, "--output_mp4_path", output_mp4_path, "--smpl_model_path", cfg.RENDER.SMPL_MODEL_PATH]
    os.system(" ".join(render_cmd))
    # subprocess.run(cmd3)
    
    return output_mp4_path, video_fname, output_npy_path, feats_fname


def load_motion(motion_uploaded, method):
    file = motion_uploaded['file']

    feats = torch.tensor(np.load(file), device=model.device)
    if len(feats.shape) == 2:
        feats = feats[None]
    # feats = model.datamodule.normalize(feats)

    # Motion tokens
    motion_lengths = feats.shape[0]
    motion_token, _ = model.vae.encode(feats)

    motion_token_string = model.lm.motion_token_to_string(
        motion_token, [motion_token.shape[1]])[0]
    motion_token_length = motion_token.shape[1]

    # Motion rendered
    joints = model.datamodule.feats2joints(feats.cpu()).cpu().numpy()
    output_mp4_path, video_fname, output_npy_path, joints_fname = render_motion(
        joints,
        feats.to('cpu').numpy(), method)

    motion_uploaded.update({
        "feats": feats,
        "joints": joints,
        "motion_video": output_mp4_path,
        "motion_video_fname": video_fname,
        "motion_joints": output_npy_path,
        "motion_joints_fname": joints_fname,
        "motion_lengths": motion_lengths,
        "motion_token": motion_token,
        "motion_token_string": motion_token_string,
        "motion_token_length": motion_token_length,
    })

    return motion_uploaded


def add_text(history, text, motion_uploaded, data_stored, method):
    data_stored = data_stored + [{'user_input': text}]

    text = f"""<h3>{text}</h3>"""
    history = history + [(text, None)]
    if 'file' in motion_uploaded.keys():
        motion_uploaded = load_motion(motion_uploaded, method)
        output_mp4_path = motion_uploaded['motion_video']
        video_fname = motion_uploaded['motion_video_fname']
        output_npy_path = motion_uploaded['motion_joints']
        joints_fname = motion_uploaded['motion_joints_fname']
        history = history + [(Video_Components.format(
            video_path=output_mp4_path,
            video_fname=video_fname,
            motion_path=output_npy_path,
            motion_fname=joints_fname), None)]

    return history, gr.update(value="",
                              interactive=False), motion_uploaded, data_stored


def add_audio(history, audio_path, data_stored, language='en'):
    audio, sampling_rate = librosa.load(audio_path, sr=16000)
    input_features = audio_processor(
        audio, sampling_rate, return_tensors="pt"
    ).input_features  # whisper training sampling rate, do not modify
    input_features = torch.Tensor(input_features).to(device)

    if language == 'English':
        forced_decoder_ids = forced_decoder_ids_en
    else:
        forced_decoder_ids = forced_decoder_ids_zh
    predicted_ids = audio_model.generate(input_features,
                                         forced_decoder_ids=forced_decoder_ids)
    text_input = audio_processor.batch_decode(predicted_ids,
                                              skip_special_tokens=True)
    text_input = str(text_input).strip('[]"')
    data_stored = data_stored + [{'user_input': text_input}]
    gr.update(value=data_stored, interactive=False)
    history = history + [(text_input, None)]

    return history, data_stored


def add_file(history, file, txt, motion_uploaded):
    motion_uploaded['file'] = file.name
    txt = txt.replace(" <Motion_Placeholder>", "") + " <Motion_Placeholder>"
    return history, gr.update(value=txt, interactive=True), motion_uploaded


def bot(history, motion_uploaded, data_stored, method):

    motion_length, motion_token_string = motion_uploaded[
        "motion_lengths"], motion_uploaded["motion_token_string"]

    input = data_stored[-1]['user_input']
    prompt = model.lm.placeholder_fulfill(input, motion_length,
                                          motion_token_string, "")
    data_stored[-1]['model_input'] = prompt
    batch = {
        "length": [motion_length],
        "text": [prompt],
    }

    outputs = model(batch, task="t2m")
    out_feats = outputs["feats"][0]
    out_lengths = outputs["length"][0]
    out_joints = outputs["joints"][:out_lengths].detach().cpu().numpy()
    out_texts = outputs["texts"][0]
    output_mp4_path, video_fname, output_npy_path, joints_fname = render_motion(
        out_joints,
        out_feats.to('cpu').numpy(), method)

    motion_uploaded = {
        "feats": None,
        "joints": None,
        "motion_video": None,
        "motion_lengths": 0,
        "motion_token": None,
        "motion_token_string": '',
        "motion_token_length": 0,
    }

    data_stored[-1]['model_output'] = {
        "feats": out_feats,
        "joints": out_joints,
        "length": out_lengths,
        "texts": out_texts,
        "motion_video": output_mp4_path,
        "motion_video_fname": video_fname,
        "motion_joints": output_npy_path,
        "motion_joints_fname": joints_fname,
    }

    if '<Motion_Placeholder>' == out_texts:
        response = [
            Video_Components.format(video_path=output_mp4_path,
                                    video_fname=video_fname,
                                    motion_path=output_npy_path,
                                    motion_fname=joints_fname)
        ]
    elif '<Motion_Placeholder>' in out_texts:
        response = [
            Text_Components.format(
                msg=out_texts.split("<Motion_Placeholder>")[0]),
            Video_Components.format(video_path=output_mp4_path,
                                    video_fname=video_fname,
                                    motion_path=output_npy_path,
                                    motion_fname=joints_fname),
            Text_Components.format(
                msg=out_texts.split("<Motion_Placeholder>")[1]),
        ]
    else:
        response = f"""<h3>{out_texts}</h3>"""

    history[-1][1] = ""
    for character in response:
        history[-1][1] += character
        time.sleep(0.02)
        yield history, motion_uploaded, data_stored


def bot_example(history, responses):
    history = history + responses
    return history


with open("assets/css/custom.css", "r", encoding="utf-8") as f:
    customCSS = f.read()

with gr.Blocks(css=customCSS) as demo:

    # Examples
    chat_instruct = gr.State([
        (None,
         "πŸ‘‹ Hi, I'm MotionGPT! I can generate realistic human motion from text, or generate text from motion."
         ),
        (None,
         "πŸ’‘ You can chat with me in pure text like generating human motion following your descriptions."
         ),
        (None,
         "πŸ’‘ After generation, you can click the button in the top right of generation human motion result to download the human motion video or feature stored in .npy format."
         ),
        (None,
         "πŸ’‘ With the human motion feature file downloaded or got from dataset, you are able to ask me to translate it!"
         ),
        (None,
         "πŸ’‘ Of courser, you can also purely chat with me and let me give you human motion in text, here are some examples!"
         ),
        (None,
         "πŸ’‘ We provide two motion visulization methods. The default fast method is skeleton line ploting which is like the examples below:"
         ),
        (None,
         Video_Components_example.format(
             video_path="assets/videos/example0_fast.mp4",
             video_fname="example0_fast.mp4")),
        (None,
         "πŸ’‘ And the slow method is SMPL model rendering which is more realistic but slower."
         ),
        (None,
         Video_Components_example.format(
             video_path="assets/videos/example0.mp4",
             video_fname="example0.mp4")),
        (None,
         "πŸ’‘ If you want to get the video in our paper and website like below, you can refer to the scirpt in our [github repo](https://github.com/OpenMotionLab/MotionGPT#-visualization)."
         ),
        (None,
         Video_Components_example.format(
             video_path="assets/videos/example0_blender.mp4",
             video_fname="example0_blender.mp4")),
        (None, "πŸ‘‰ Follow the examples and try yourself!"),
    ])
    chat_instruct_sum = gr.State([(None, '''
         πŸ‘‹ Hi, I'm MotionGPT! I can generate realistic human motion from text, or generate text from motion.
         
         1. You can chat with me in pure text like generating human motion following your descriptions.
         2. After generation, you can click the button in the top right of generation human motion result to download the human motion video or feature stored in .npy format.
         3. With the human motion feature file downloaded or got from dataset, you are able to ask me to translate it!
         4. Of course, you can also purely chat with me and let me give you human motion in text, here are some examples!
         ''')] + chat_instruct.value[-7:])

    t2m_examples = gr.State([
        (None,
         "πŸ’‘ You can chat with me in pure text, following are some examples of text-to-motion generation!"
         ),
        ("A person is walking forwards, but stumbles and steps back, then carries on forward.",
         Video_Components_example.format(
             video_path="assets/videos/example0.mp4",
             video_fname="example0.mp4")),
        ("Generate a man aggressively kicks an object to the left using his right foot.",
         Video_Components_example.format(
             video_path="assets/videos/example1.mp4",
             video_fname="example1.mp4")),
        ("Generate a person lowers their arms, gets onto all fours, and crawls.",
         Video_Components_example.format(
             video_path="assets/videos/example2.mp4",
             video_fname="example2.mp4")),
        ("Show me the video of a person bends over and picks things up with both hands individually, then walks forward.",
         Video_Components_example.format(
             video_path="assets/videos/example3.mp4",
             video_fname="example3.mp4")),
        ("Imagine a person is practing balancing on one leg.",
         Video_Components_example.format(
             video_path="assets/videos/example5.mp4",
             video_fname="example5.mp4")),
        ("Show me a person walks forward, stops, turns directly to their right, then walks forward again.",
         Video_Components_example.format(
             video_path="assets/videos/example6.mp4",
             video_fname="example6.mp4")),
        ("I saw a person sits on the ledge of something then gets off and walks away.",
         Video_Components_example.format(
             video_path="assets/videos/example7.mp4",
             video_fname="example7.mp4")),
        ("Show me a person is crouched down and walking around sneakily.",
         Video_Components_example.format(
             video_path="assets/videos/example8.mp4",
             video_fname="example8.mp4")),
    ])

    m2t_examples = gr.State([
        (None,
         "πŸ’‘ With the human motion feature file downloaded or got from dataset, you are able to ask me to translate it, here are some examples!"
         ),
        ("Please explain the movement shown in <Motion_Placeholder> using natural language.",
         None),
        (Video_Components_example.format(
            video_path="assets/videos/example0.mp4",
            video_fname="example0.mp4"),
         "The person was pushed but didn't fall down"),
        ("What kind of action is being represented in <Motion_Placeholder>? Explain it in text.",
         None),
        (Video_Components_example.format(
            video_path="assets/videos/example4.mp4",
            video_fname="example4.mp4"),
         "The figure has its hands curled at jaw level, steps onto its left foot and raises right leg with bent knee to kick forward and return to starting stance."
         ),
        ("Provide a summary of the motion demonstrated in <Motion_Placeholder> using words.",
         None),
        (Video_Components_example.format(
            video_path="assets/videos/example2.mp4",
            video_fname="example2.mp4"),
         "A person who is standing with his arms up and away from his sides bends over, gets down on his hands and then his knees and crawls forward."
         ),
        ("Generate text for <Motion_Placeholder>:", None),
        (Video_Components_example.format(
            video_path="assets/videos/example5.mp4",
            video_fname="example5.mp4"),
         "The man tries to stand in a yoga tree pose and looses his balance."),
        ("Provide a summary of the motion depicted in <Motion_Placeholder> using language.",
         None),
        (Video_Components_example.format(
            video_path="assets/videos/example6.mp4",
            video_fname="example6.mp4"),
         "Person walks up some steps then leeps to the other side and goes up a few more steps and jumps dow"
         ),
        ("Describe the motion represented by <Motion_Placeholder> in plain English.",
         None),
        (Video_Components_example.format(
            video_path="assets/videos/example7.mp4",
            video_fname="example7.mp4"),
         "Person sits down, then stands up and walks forward. then the turns around 180 degrees and walks the opposite direction"
         ),
        ("Provide a description of the action in <Motion_Placeholder> using words.",
         None),
        (Video_Components_example.format(
            video_path="assets/videos/example8.mp4",
            video_fname="example8.mp4"),
         "This man is bent forward and walks slowly around."),
    ])

    t2t_examples = gr.State([
        (None,
         "πŸ’‘ Of course, you can also purely chat with me and let me give you human motion in text, here are some examples!"
         ),
        ('Depict a motion as like you have seen it.',
         "A person slowly walked forward in rigth direction while making the circle"
         ),
        ('Random say something about describing a human motion.',
         "A man throws punches using his right hand."),
        ('Describe the motion of someone as you will.',
         "Person is moving left to right in a dancing stance swaying hips, moving feet left to right with arms held out"
         ),
        ('Come up with a human motion caption.',
         "A person is walking in a counter counterclockwise motion."),
        ('Write a sentence about how someone might dance.',
         "A person with his hands down by his sides reaches down for something with his right hand, uses the object to make a stirring motion, then places the item back down."
         ),
        ('Depict a motion as like you have seen it.',
         "A person is walking forward a few feet, then turns around, walks back, and continues walking."
         )
    ])

    Init_chatbot = chat_instruct.value[:
                                       1] + t2m_examples.value[:
                                                               3] + m2t_examples.value[:3] + t2t_examples.value[:2] + chat_instruct.value[
                                                                   -7:]

    # Variables
    motion_uploaded = gr.State({
        "feats": None,
        "joints": None,
        "motion_video": None,
        "motion_lengths": 0,
        "motion_token": None,
        "motion_token_string": '',
        "motion_token_length": 0,
    })
    data_stored = gr.State([])

    gr.Markdown('''
# MotionGPT: Human Motion as a Foreign Language

<div>
<a style="display:inline-block" href="https://https://motion-gpt.github.io/"><img src='https://img.shields.io/badge/project_page-9375A7'></a>
<a style="display:inline-block; margin-left: .5em" href="https://arxiv.org/abs/2306.14795"><img src="https://img.shields.io/badge/2309.16653-f9f7f7?logo=data:image/png;base64,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"></a>
<a style="display:inline-block; margin-left: .5em" href='https://github.com/OpenMotionLab/MotionGPT'><img src='https://img.shields.io/github/stars/OpenMotionLab/MotionGPT?style=social'/></a>
<a style="display:inline-block; margin-left: .5em" href="https://github.com/OpenMotionLab/MotionGPT#-citation"><img src="https://img.shields.io/badge/Citation-4385FE?&logo=google-scholar&logoColor=white" alt="Citation"> </a>
</div>
            
                ''')

    chatbot = gr.Chatbot(Init_chatbot,
                         elem_id="mGPT",
                         height=600,
                         label="MotionGPT",
                         avatar_images=(None,
                                        ("assets/images/avatar_bot.jpg")),
                         bubble_full_width=False)

    with gr.Row():
        with gr.Column(scale=0.85):
            with gr.Row():
                txt = gr.Textbox(
                    label="Text",
                    show_label=False,
                    elem_id="textbox",
                    placeholder=
                    "Enter text and press ENTER or speak to input. You can also upload motion.",
                    container=False)

            with gr.Row():
                aud = gr.Audio(source="microphone",
                               label="Speak input",
                               type='filepath')
                btn = gr.UploadButton("πŸ“ Upload motion",
                                      elem_id="upload",
                                      file_types=["file"])
                # regen = gr.Button("πŸ”„ Regenerate", elem_id="regen")
                clear = gr.ClearButton([txt, chatbot, aud], value='πŸ—‘οΈ Clear')

            with gr.Row():
                gr.Markdown('''
                ### You can get more examples (pre-generated for faster response) by clicking the buttons below:
                ''')

            with gr.Row():
                instruct_eg = gr.Button("Instructions", elem_id="instruct")
                t2m_eg = gr.Button("Text-to-Motion", elem_id="t2m")
                m2t_eg = gr.Button("Motion-to-Text", elem_id="m2t")
                t2t_eg = gr.Button("Random description", elem_id="t2t")

        with gr.Column(scale=0.15, min_width=150):
            method = gr.Dropdown(["slow", "fast"],
                                 label="Visulization method",
                                 interactive=True,
                                 elem_id="method",
                                 value="slow")

            language = gr.Dropdown(["English", "δΈ­ζ–‡"],
                                   label="Speech language",
                                   interactive=True,
                                   elem_id="language",
                                   value="English")

    txt_msg = txt.submit(
        add_text, [chatbot, txt, motion_uploaded, data_stored, method],
        [chatbot, txt, motion_uploaded, data_stored],
        queue=False).then(bot, [chatbot, motion_uploaded, data_stored, method],
                          [chatbot, motion_uploaded, data_stored])

    txt_msg.then(lambda: gr.update(interactive=True), None, [txt], queue=False)

    file_msg = btn.upload(add_file, [chatbot, btn, txt, motion_uploaded],
                          [chatbot, txt, motion_uploaded],
                          queue=False)
    aud_msg = aud.stop_recording(
        add_audio, [chatbot, aud, data_stored, language],
        [chatbot, data_stored],
        queue=False).then(bot, [chatbot, motion_uploaded, data_stored, method],
                          [chatbot, motion_uploaded, data_stored])
    # regen_msg = regen.click(bot,
    #                         [chatbot, motion_uploaded, data_stored, method],
    #                         [chatbot, motion_uploaded, data_stored],
    #                         queue=False)

    instruct_msg = instruct_eg.click(bot_example, [chatbot, chat_instruct_sum],
                                     [chatbot],
                                     queue=False)
    t2m_eg_msg = t2m_eg.click(bot_example, [chatbot, t2m_examples], [chatbot],
                              queue=False)
    m2t_eg_msg = m2t_eg.click(bot_example, [chatbot, m2t_examples], [chatbot],
                              queue=False)
    t2t_eg_msg = t2t_eg.click(bot_example, [chatbot, t2t_examples], [chatbot],
                              queue=False)

    chatbot.change(scroll_to_output=True)

if __name__ == "__main__":
    demo.launch(debug=True)