-
Notifications
You must be signed in to change notification settings - Fork 35
/
app.py
310 lines (264 loc) · 12.3 KB
/
app.py
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
import argparse
import os
import random
# import sys
# import os
#
# BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
# sys.path.append(BASE_DIR)
# os.system("CUDA_HOME=/usr/local/cuda python -m pip uninstall groundingdino")
os.system("git clone https://github.com/IDEA-Research/GroundingDINO.git"
"&& export CUDA_HOME=/usr/local/cuda; pip install -e GroundingDINO")
import sys
sys.path.insert(0, './GroundingDINO')
import numpy as np
import torch
import torch.backends.cudnn as cudnn
import gradio as gr
from constants.constant import LIGHTER_COLOR_MAP_HEX
# NOTE: Must import LlamaTokenizer before `bubogpt.common.config`
# otherwise, it will cause seg fault when `llama_tokenizer.decode` is called
from grounding_model import GroundingModule
from match import MatchModule
from bubogpt.common.config import Config
from bubogpt.common.dist_utils import get_rank
from bubogpt.common.registry import registry
from eval_scripts.conversation import Chat, CONV_X, DummyChat
# NOTE&TODO: put this before bubogpt import will cause circular import
# possibly because `imagebind` imports `bubogpt` and `bubogpt` also imports `imagebind`
from imagebind.models.image_bind import ModalityType
# from ner import NERModule
from tagging_model import TaggingModule
def parse_args():
parser = argparse.ArgumentParser(description="Qualitative")
parser.add_argument("--cfg-path", help="path to configuration file.", default='./eval_configs/mmgpt4_eval.yaml')
parser.add_argument("--dummy", action="store_true", help="Debug Mode")
parser.add_argument("--gpu-id", type=int, default=0, help="specify the gpu to load the model.")
parser.add_argument(
"--options",
nargs="+",
help="override some settings in the used config, the key-value pair "
"in xxx=yyy format will be merged into config file (deprecate), "
"change to --cfg-options instead.",
)
parser.add_argument("--ground-all", action="store_true")
args = parser.parse_args()
return args
def setup_seeds(config):
seed = config.run_cfg.seed + get_rank()
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
cudnn.benchmark = False
cudnn.deterministic = True
# ========================================
# Model Initialization
# ========================================
print('Initializing Chat')
args = parse_args()
assert args.dummy or (args.cfg_path is not None), "Invalid Config! Set --dummy or configurate the cfg_path!"
device = 'cuda'.format(args.gpu_id) if torch.cuda.is_available() else 'cpu'
if not args.dummy:
cfg = Config(args)
# Create processors
vis_processor_cfg = cfg.datasets_cfg.default.vis_processor.eval
aud_processor_cfg = cfg.datasets_cfg.default.audio_processor.eval
vis_processor = registry.get_processor_class(vis_processor_cfg.name).from_config(vis_processor_cfg)
aud_processor = registry.get_processor_class(aud_processor_cfg.name).from_config(aud_processor_cfg)
processors = {ModalityType.VISION: vis_processor, ModalityType.AUDIO: aud_processor}
# Create model
model_config = cfg.model_cfg
model_config.device_8bit = device
model_cls = registry.get_model_class(model_config.arch)
model = model_cls.from_config(model_config).to(device)
chat = Chat(model, processors, device=device)
else:
model = None
chat = DummyChat()
match = MatchModule(model='gpt-3.5-turbo')
tagging_module = TaggingModule(device=device)
grounding_dino = GroundingModule(device=device)
print('Initialization Finished')
# ========================================
# Gradio Setting
# ========================================
def gradio_reset(chat_state, emb_list):
if chat_state is not None:
chat_state.messages = []
if emb_list is not None:
emb_list = []
return None, gr.update(value=None, interactive=True), gr.update(value=None, interactive=False), \
gr.update(value=None, interactive=True), \
gr.update(placeholder='Please upload your image/audio first', interactive=False), \
gr.update(value=None), \
gr.update(value="Upload & Start Chat", interactive=True), \
chat_state, emb_list, gr.update(value={})
def upload_x(gr_img, gr_aud, chat_state):
if gr_img is None and gr_aud is None:
return None, None, None, gr.update(interactive=True), chat_state, None, {}
chat_state = CONV_X.copy()
emb_list = []
if gr_img is not None:
chat.upload_img(gr_img, chat_state, emb_list)
state = {
'tags': tagging_module(gr_img)
}
# print(state)
else:
state = {}
if gr_aud is not None:
chat.upload_aud(gr_aud, chat_state, emb_list)
return gr.update(interactive=False), gr.update(interactive=False), \
gr.update(interactive=True, placeholder='Type and press Enter'), \
gr.update(value="Start Chatting", interactive=False), \
chat_state, emb_list, state
def gradio_ask(user_message, chatbot, chat_state, text_output, last_answer):
if len(user_message) == 0:
return gr.update(interactive=True, placeholder='Input should not be empty!'), chatbot, chat_state, \
gr.update(value=None, color_map=None, show_legend=False), gr.update(value=None)
if last_answer is not None:
chatbot[-1][1] = last_answer
chat.ask(user_message, chat_state)
if text_output is not None:
os.makedirs('results', exist_ok=True)
# print("****** Text output is:", text_output)
chatbot[-1][1] = ''.join(map(lambda x: x[0], text_output))
chatbot = chatbot + [[user_message, None]]
return '', chatbot, chat_state, gr.update(value=None, color_map=None, show_legend=False), gr.update(value=None)
def gradio_answer(image, chatbot, chat_state, emb_list, num_beams, temperature, entity_state):
llm_message = chat.answer(conversation=chat_state,
emb_list=emb_list,
num_beams=num_beams,
temperature=temperature,
max_new_tokens=300,
max_length=2000)[0]
if image is not None:
# new_entity_state = entity_state.value()
# new_entity_state.update({"answer": llm_message})
entity_state["answer"] = llm_message
rich_text, match_state, color_map = match(llm_message, entity_state)
print("Original Color Map: ", color_map)
color_map = {key: LIGHTER_COLOR_MAP_HEX[color_map[key]] for key in color_map}
print("Modified Color Map: ", color_map)
chatbot[-1][1] = "The answer can be found in the textbox below and I'm trying my best to highlight the " \
"corresponding region on the image."
# new_entity_state.update({"match_state": match_state})
entity_state['match_state'] = match_state # item_id -> local_id
new_grounded_image = grounding_dino.draw(image, entity_state)
show_legend = bool(match_state)
# print('gradio_answer ==> current state: ', entity_state)
# if args.ground_all:
# ground_img, local_results = grounding_dino.prompt2mask(image,
# '.'.join(map(lambda x: x, state['entity'])),
# state=state)
# else:
# ground_img = None
return chatbot, chat_state, emb_list, \
gr.update(value=rich_text, color_map=color_map, show_legend=show_legend), \
gr.update(value=entity_state), \
gr.update(value=llm_message), gr.update(value=new_grounded_image)
else:
chatbot[-1][1] = llm_message
return chatbot, chat_state, emb_list, \
gr.update(value=None), \
entity_state, \
gr.update(value=None), gr.update(value=None)
def grounding_fn(image, chatbot, entity_state):
# print("Grounding fn: ", entity_state)
if image and entity_state:
ground_img, local_results = grounding_dino.prompt2mask2(
image, ','.join(map(lambda x: x, entity_state['tags'])), state=entity_state
)
entity_state['grounding'] = {
'full': ground_img,
'local': local_results
}
# print('grounding_fn ==> current state: ', entity_state)
return chatbot, gr.update(value=ground_img, interactive=False), entity_state
return chatbot, gr.update(value=None, interactive=False), entity_state
def select_fn(image, ground_img, entity_state, evt: gr.SelectData):
if image is None:
return gr.update(value=None, interactive=False)
item, label = evt.value[0], evt.value[1]
if label is None:
return ground_img
# print('select_fn ==> current state: ', entity_state)
torch.cuda.synchronize()
if 'grounding' not in entity_state:
ground_img, local_results = grounding_dino.prompt2mask2(image,
','.join(map(lambda x: x[0], entity_state['tags'])),
state=entity_state)
entity_state['grounding'] = {
'full': ground_img,
'local': local_results
}
# local_img = entity_state['grounding']['local'][entity]['image']
# print("DEBUG INFO: ", entity_state)
local_img = grounding_dino.draw(image, entity_state, item.lower())
return gr.update(value=local_img, interactive=False)
title = """<h1 align="center">Demo of BuboGPT</h1>"""
description = """<h3>This is the demo of BuboGPT. Upload and start chatting!</h3>"""
# article = """<p><a href='https://minigpt-4.github.io'><img src='https://img.shields.io/badge/Project-Page-Green'></a></p><p><a href='https://github.com/Vision-CAIR/MiniGPT-4'><img src='https://img.shields.io/badge/Github-Code-blue'></a></p><p><a href='https://raw.githubusercontent.com/Vision-CAIR/MiniGPT-4/main/MiniGPT_4.pdf'><img src='https://img.shields.io/badge/Paper-PDF-red'></a></p>
# """
# TODO show examples below
with gr.Blocks() as demo:
gr.Markdown(title)
gr.Markdown(description)
# gr.Markdown(article)
with gr.Row():
with gr.Column(scale=0.5):
image = gr.Image(type="pil")
grounded_image = gr.Image(type="pil", interactive=False)
audio = gr.Audio()
upload_button = gr.Button(value="Upload & Start Chat", interactive=True, variant="primary")
clear = gr.Button("Restart")
num_beams = gr.Slider(
minimum=1,
maximum=10,
value=1,
step=1,
interactive=True,
label="beam search numbers",
)
temperature = gr.Slider(
minimum=0.1,
maximum=2.0,
value=1.0,
step=0.1,
interactive=True,
label="Temperature",
)
with gr.Column():
chat_state = gr.State()
last_answer = gr.State()
entity_state = gr.State(value={})
emb_list = gr.State()
chatbot = gr.Chatbot(label='BuboGPT')
text_output = gr.HighlightedText(value=None, label="Response", show_legend=False)
text_input = gr.Textbox(label='User', placeholder='Please upload your image/audio first', interactive=False)
upload_button.click(
upload_x, [image, audio, chat_state],
[image, audio, text_input, upload_button, chat_state, emb_list, entity_state]).then(
grounding_fn,
[image, chatbot, entity_state],
[chatbot, grounded_image, entity_state]
)
text_input.submit(gradio_ask,
[text_input, chatbot, chat_state, text_output, last_answer],
[text_input, chatbot, chat_state, text_output, last_answer]
).then(
gradio_answer,
[image, chatbot, chat_state, emb_list, num_beams, temperature, entity_state],
[chatbot, chat_state, emb_list, text_output, entity_state, last_answer, grounded_image]
)
clear.click(gradio_reset,
[chat_state, emb_list],
[chatbot, image, grounded_image, audio, text_input, text_output,
upload_button, chat_state, emb_list, entity_state],
queue=False)
text_output.select(
select_fn,
[image, grounded_image, entity_state],
[grounded_image]
)
demo.launch(enable_queue=True)