File size: 11,111 Bytes
915477c c58a0c0 915477c 40b9fa3 915477c |
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 |
import os
import re
import json
import streamlit as st
from PIL import Image, ImageDraw
import requests
from io import BytesIO
import seaborn as sns
import matplotlib.pyplot as plt
from streamlit_chat import message as st_message
import yaml
st.set_page_config(page_title="Data Exploration", page_icon="🌍", layout="wide", initial_sidebar_state="collapsed")
COLORS = sns.color_palette("Paired", n_colors=100).as_hex()
def load_config(config_fn, field='data_explore') -> dict:
config = yaml.load(open(config_fn), Loader=yaml.Loader)
return config[field]
def convert_from_prompt_tokens(s_with_region_tokens):
"""Convert from strings with prompt tokens for prompt encoders
e.g.:
Input: "<Region><L12><24><L101><L777></Region>"
Output: [0.012, 0.024, 0.101, 0.777]
"""
REGION_PATTERN = r'<Region>(\s*<L(\d{1,4})>\s*<L(\d{1,4})>\s*<L(\d{1,4})>\s*<L(\d{1,4})>\s*)</Region>'
boxes = []
boxes_str = re.findall(REGION_PATTERN, s_with_region_tokens)
for boxes_str_i in boxes_str:
matched_str_i, boxes_str_i = boxes_str_i[0], boxes_str_i[1:]
boxes.append(tuple([int(s)/1000 for s in boxes_str_i]))
return boxes
def parse_regions(s):
pattern = r"\[([\d.,\s]+)\]"
matches = re.findall(pattern, s)
bboxes = []
points = []
for res in matches:
res = eval(res)
if len(res) == 4:
# bbox
x1, y1, x2, y2 = res
bboxes.append((x1, y1, x2, y2))
else:
x1, y1 = res
points.append((x1, y1))
bboxes.extend(convert_from_prompt_tokens(s))
return list(set(bboxes))
def get_image(image_path, bboxes=None):
if os.path.exists(image_path):
image = Image.open(image_path).convert('RGB')
else:
# 从URL获取图片
response = requests.get(image_path)
image = Image.open(BytesIO(response.content)).convert('RGB')
draw = ImageDraw.Draw(image, 'RGB')
color_mapping = None
if bboxes is not None:
width, height = image.size
color_mapping = []
for i, bbox_coords in enumerate(bboxes):
color = COLORS[i]
x1, y1, x2, y2 = bbox_coords
x1 *= width
y1 *= height
x2 *= width
y2 *= height
draw.rectangle([x1, y1, x2, y2], outline=color, width=3)
color_mapping.append([bbox_coords, color])
color_mapping = dict(color_mapping)
return image, color_mapping
def insert_color(s, color_mapping):
for coords, color in color_mapping.items():
coords_str = ', '.join([str(x) for x in coords])
s = s.replace('[' + coords_str + ']', f'<span style="color: {color}; font-weight: bold;">■</span>' + ' [' + coords_str + ']')
return s
modal_indicator = ['<image>', '<audio>', '<video>']
def show_one_msg(msg, modal_inputs):
splits = re.split('(' + '|'.join(modal_indicator) + ')', msg)
for s in splits:
if s == '<image>':
st.image(modal_inputs['image'].pop(0))
elif s == '<audio>':
st.audio(modal_inputs['audio'].pop(0))
elif s == '<video>':
st.video(modal_inputs['video'].pop(0))
else:
st.write(s)
def show_multimodal_example(example, col1, col2):
with col1:
info = example.get('info', {})
info['modal_inputs'] = example['modal_inputs']
st.json(info)
with col2:
conversations = example['conversations']
modal_inputs = example['modal_inputs']
for i in range(len(conversations) // 2):
with st.chat_message("user"):
show_one_msg(conversations[2*i]['value'], modal_inputs)
with st.chat_message("assistant"):
show_one_msg(conversations[2*i+1]['value'], modal_inputs)
def show_example(example, col1, col2, enable_scores=True):
if 'conversations' in example:
regions = parse_regions(str(example['conversations']))
else:
regions = parse_regions(str(example))
image_fn = example['image']
image, color_mapping = get_image(image_fn, regions)
with col1:
st.image(image)
info = example.get('info', {})
info['id'] = example.get('id', 'N/A')
info['image'] = image_fn
if 'dataset' in example:
info['source'] = example['dataset']
st.json(info)
if len(color_mapping):
table_md = "| 颜色 | 坐标 |\n| --- | --- |\n"
for coords, color in color_mapping.items():
color_cell = f'<span style="color: {color}; font-weight: bold;">■</span>'
table_md += f"| {color_cell} | {coords} |\n"
# 使用Markdown显示表格
st.markdown(table_md, unsafe_allow_html=True)
score_dict = None
with col2:
if 'conversations' in example:
if enable_scores:
score_dict = {'image': image_fn, 'conversations': example['conversations']}
with st.expander("Give a score based on the result above", expanded=True):
quality_score = st.radio("问题质量分数",('Bad', 'Mediocre', 'Good'),key="quality", horizontal = True)
format_score = st.radio("格式分数",('Bad', 'Mediocre', 'Good'),key="format", horizontal = True)
score_dict['scores'] = {
'quality': quality_score, 'format': format_score
}
st.subheader("Chat")
conversations = example['conversations']
for i in range(len(conversations) // 2):
st_message(conversations[2*i]['value'], is_user=True, key=image_fn + str(2*i))
st_message(conversations[2*i+1]['value'], is_user=False, key=image_fn + str(2*i+1))
if 'ground_truth' in example:
# 显示查询
gt = insert_color(json.dumps(example['ground_truth']), color_mapping)
st.markdown(f"**Ground Truth:**\n\n{gt}", unsafe_allow_html=True)
else:
# 显示指令
instruction = insert_color(example['instruction'], color_mapping)
st.markdown(f"**Instruction:**\n\n{instruction}", unsafe_allow_html=True)
# 显示输入
if 'input' in example:
input = insert_color(example['input'], color_mapping)
st.markdown(f"**Input:**\n\n{input}", unsafe_allow_html=True)
# 显示输出
output = insert_color(example['output'], color_mapping)
st.markdown(f"**Output:**\n\n{output}", unsafe_allow_html=True)
if 'query' in example:
# 显示查询
query = insert_color(json.dumps(example['query']), color_mapping)
st.markdown(f"**Query:**\n\n{query}", unsafe_allow_html=True)
return score_dict
def reset_state():
print('RESET')
st.session_state['data_explore'] = {'idx': 0}
st.session_state.scores = {}
def load_dir_data(dir, dataset_configs):
mapping_file = os.path.join(dir, 'mapping.yaml')
assert os.path.exists(mapping_file)
config = yaml.load(open(mapping_file), Loader=yaml.Loader)
# image_paths = dataset_configs
image_paths = config['image_paths']
image_paths['default'] = image_paths.get('default', '.')
res = []
for k, v in config['mapping'].items():
if os.path.exists(os.path.join(dir, k + '.json')):
data = json.load(open(os.path.join(dir, k + '.json')))
elif os.path.exists(os.path.join(dir, k + '.jsonl')):
data = [json.loads(line) for line in open(os.path.join(dir, k + '.jsonl'))]
elif os.path.exists(os.path.join(dir, k + '.txt')):
data = [json.loads(line) for line in open(os.path.join(dir, k + '.txt'))]
image_path = image_paths.get(v, image_paths['default'])
for example in data:
example['image'] = os.path.join(image_path, example['image'])
example['dataset'] = k
res.extend(data)
return res
@st.cache_data
def load_data(fn, dataset_configs):
if os.path.isdir(fn):
res = load_dir_data(fn, dataset_configs)
return res
if fn.endswith(('.txt', '.jsonl')):
res = []
for line in open(fn):
example = json.loads(line)
res.append(example)
else:
res = json.load(open(fn))
for example in res:
dataset_path = dataset_configs[example.get('dataset', 'default')]
if 'image' in example:
example['image'] = os.path.join(dataset_path, example['image'])
elif 'img_info' in example:
if isinstance(example['img_info'], str):
example['image'] = os.path.join(dataset_path, example['img_info'])
else:
if 'coco_url' in example['img_info']:
example['image'] = example['img_info']['coco_url']
else:
assert 'modal_inputs' in example
return res
dataset_configs = load_config('config.yaml')
print(dataset_configs)
data_paths = dataset_configs.get('data_paths', ['instruction_data'])
files = []
def add_file(path):
if os.path.exists(os.path.join(path, 'mapping.yaml')):
files.append(path)
else:
for f in sorted(os.listdir(path)):
file = os.path.join(path, f)
if os.path.isfile(file) and file.endswith(('.txt', '.json')):
files.append(file)
else:
add_file(file)
for data_path in data_paths:
add_file(data_path)
st.session_state['data_explore'] = {'idx': 0}
enable_score = st.sidebar.checkbox('Score it!', value=False)
if enable_score and 'scores' not in st.session_state:
st.session_state.scores = {}
status_placeholder = st.empty()
control_col1, control_col2 = st.columns(2)
with control_col1:
selected_file = st.selectbox('Select a file', files, on_change=reset_state)
col1, col2 = st.columns(2)
if selected_file:
data = load_data(selected_file, dataset_configs)
with control_col2:
idx = st.number_input(f'Input an idx (Total: {len(data)})', min_value=0, max_value=len(data), value=st.session_state.get('data_explore', {}).get('idx', 0))
st.session_state['data_explore']['idx'] = idx
if 'image' in data[idx]:
show_example(data[idx], col1, col2, enable_scores=enable_score)
else:
show_multimodal_example(data[idx], col1, col2)
if enable_score:
name = st.sidebar.text_input("Username", placeholder = "Enter your name", value="cc")
if st.sidebar.button(label ="Submit scores", key = "submit"):
if name:
score_path = f"score_results/{os.path.basename(selected_file)}_{name}.json"
with open(score_path, "w") as score_file:
json.dump(st.session_state.scores, score_file, indent = 4)
status_placeholder.success("Successfully saved!")
else:
status_placeholder.error("Please enter your name on the sidebar!") |