-
Notifications
You must be signed in to change notification settings - Fork 15
/
model_seg_loader.py
185 lines (155 loc) · 8.45 KB
/
model_seg_loader.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
import argparse
import torch
import os
import json
from tqdm import tqdm
import shortuuid
import random
import glob
from vcoder_llava.constants import (
IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN,
SEG_TOKEN_INDEX, DEFAULT_SEG_TOKEN,
)
from vcoder_llava.vcoder_conversation import conv_templates, SeparatorStyle
from vcoder_llava.model.builder import load_pretrained_model
from vcoder_llava.utils import disable_torch_init
from vcoder_llava.mm_utils import process_images, tokenizer_seg_token, tokenizer_image_token, get_model_name_from_path, KeywordsStoppingCriteria
from torch.utils.data import Dataset, DataLoader
from vcoder_llava.questions import QUESTIONS
import math
from PIL import Image
def split_list(lst, n):
"""Split a list into n (roughly) equal-sized chunks"""
chunk_size = math.ceil(len(lst) / n) # integer division
return [lst[i:i+chunk_size] for i in range(0, len(lst), chunk_size)]
def get_chunk(lst, n, k):
chunks = split_list(lst, n)
return chunks[k]
# Custom dataset class
class CustomDataset(Dataset):
def __init__(self, questions, args, seg_image_folder, tokenizer, image_processor, seg_image_processor, model_config):
self.questions = questions
self.image_folder = args.image_folder
self.seg_image_folder = seg_image_folder
self.images = glob.glob(os.path.join(args.image_folder, '*.jpg'))
self.images = get_chunk(self.images, args.num_chunks, args.chunk_idx)
if seg_image_folder is not None:
self.seg_images = glob.glob(os.path.join(seg_image_folder, '*.jpg'))
self.seg_images = get_chunk(self.seg_images, args.num_chunks, args.chunk_idx)
assert len(self.images) == len(self.seg_images), f"Number of images ({len(self.images)}) and seg images ({len(self.seg_images)}) must be the same"
else:
self.seg_images = None
self.tokenizer = tokenizer
self.image_processor = image_processor
self.seg_image_processor = seg_image_processor
self.model_config = model_config
def __getitem__(self, index):
image_file = self.images[index]
if self.seg_images is not None:
seg_image_file = self.seg_images[index]
else:
seg_image_file = None
ques = random.choice(self.questions)
qs = DEFAULT_IMAGE_TOKEN + '\n' + ques
image = Image.open(os.path.join(image_file)).convert('RGB')
image_tensor = process_images([image], self.image_processor, self.model_config)[0]
if seg_image_file is not None:
seg_image = Image.open(os.path.join(seg_image_file)).convert('RGB')
seg_image_tensor = process_images([seg_image], self.seg_image_processor, self.model_config)[0]
qs = DEFAULT_SEG_TOKEN + '\n' + qs
else:
seg_image_tensor = image_tensor
qs = qs + " Return the answer in the paragraph format: 'The objects present in the image are: ...' and then list the objects with their count in word format (if greater than 1) in front of them, like 'two people'."
conv = conv_templates[args.conv_mode].copy()
conv.append_message(conv.roles[0], qs)
conv.append_message(conv.roles[1], None)
prompt = conv.get_prompt()
if seg_image_file is None:
input_ids = tokenizer_image_token(prompt, self.tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt')
else:
input_ids = tokenizer_seg_token(prompt, self.tokenizer, IMAGE_TOKEN_INDEX, SEG_TOKEN_INDEX, return_tensors='pt')
return input_ids, image_tensor, seg_image_tensor, image_file.split("/")[-1], ques
def __len__(self):
return len(self.images)
# DataLoader
def create_data_loader(questions, args, seg_image_folder, tokenizer, image_processor, seg_image_processor, model_config, batch_size=1, num_workers=4):
assert batch_size == 1, "batch_size must be 1"
dataset = CustomDataset(questions, args, seg_image_folder, tokenizer, image_processor, seg_image_processor, model_config)
data_loader = DataLoader(dataset, batch_size=batch_size, num_workers=num_workers, shuffle=False)
return data_loader
def eval_model(args, task):
# Model
disable_torch_init()
model_path = os.path.expanduser(args.model_path)
model_name = get_model_name_from_path(model_path)
tokenizer, model, image_processor, seg_image_processor, _, context_len = load_pretrained_model(model_path, args.model_base, model_name)
questions = QUESTIONS[task]
answers_file = os.path.expanduser(args.output_file)
os.makedirs(os.path.dirname(answers_file), exist_ok=True)
answers_file = answers_file + f'_{task}_{args.num_chunks}_{args.chunk_idx}.txt'
if 'plain' in model_name and 'finetune' not in model_name.lower() and 'mmtag' not in args.conv_mode:
args.conv_mode = args.conv_mode + '_mmtag'
print(f'It seems that this is a plain model, but it is not using a mmtag prompt, auto switching to {args.conv_mode}.')
if not args.use_seg:
seg_image_folder = None
else:
seg_image_folder = os.path.join(args.seg_image_folder, f'{task}_inference')
data_loader = create_data_loader(questions, args, seg_image_folder, tokenizer, image_processor, seg_image_processor, model.config)
for input_ids, image_tensor, seg_image_tensor, image_file, ques in tqdm(data_loader, total=len(data_loader), desc=f'Generating {task} answers...'):
stop_str = conv_templates[args.conv_mode].sep if conv_templates[args.conv_mode].sep_style != SeparatorStyle.TWO else conv_templates[args.conv_mode].sep2
input_ids = input_ids.to(device='cuda', non_blocking=True)
with torch.inference_mode():
if "vcoder" in args.model_path:
output_ids = model.generate(
input_ids,
images=image_tensor.to(dtype=torch.float16, device='cuda', non_blocking=True),
segs=seg_image_tensor.to(dtype=torch.float16, device='cuda', non_blocking=True),
depths=None,
do_sample=True if args.temperature > 0 else False,
temperature=args.temperature,
top_p=args.top_p,
num_beams=args.num_beams,
max_new_tokens=512,
use_cache=True)
else:
output_ids = model.generate(
input_ids,
images=image_tensor.to(dtype=torch.float16, device='cuda', non_blocking=True),
do_sample=True if args.temperature > 0 else False,
temperature=args.temperature,
top_p=args.top_p,
num_beams=args.num_beams,
max_new_tokens=512,
use_cache=True)
input_token_len = input_ids.shape[1]
n_diff_input_output = (input_ids != output_ids[:, :input_token_len]).sum().item()
if n_diff_input_output > 0:
print(f'[Warning] {n_diff_input_output} output_ids are not the same as the input_ids')
outputs = tokenizer.batch_decode(output_ids[:, input_token_len:], skip_special_tokens=True)[0]
outputs = outputs.strip()
if outputs.endswith(stop_str):
outputs = outputs[:-len(stop_str)]
outputs = outputs.strip()
outputs = outputs.strip('\n')
with open(f'{answers_file}', 'a') as f:
f.write(f'Image: {image_file[0]}\n')
f.write(f'<<QUESTION>>: {ques[0]}\n')
f.write(f'<<ANSWER>>: {outputs}\n')
f.write('-------------------------------------------------------\n')
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--model-path", type=str, default="facebook/opt-350m")
parser.add_argument("--model-base", type=str, default=None)
parser.add_argument("--image-folder", type=str, default="")
parser.add_argument("--use_seg", action="store_true")
parser.add_argument("--seg-image-folder", type=str, default="")
parser.add_argument("--output-file", type=str, default="output")
parser.add_argument("--conv-mode", type=str, default="llava_v1")
parser.add_argument("--num-chunks", type=int, default=1)
parser.add_argument("--chunk-idx", type=int, default=0)
parser.add_argument("--temperature", type=float, default=0.2)
parser.add_argument("--top_p", type=float, default=None)
parser.add_argument("--num_beams", type=int, default=1)
args = parser.parse_args()
for task in ["semantic", "instance", "panoptic"]:
eval_model(args, task)