Spaces:
Sleeping
Sleeping
add distabled chat message
Browse files- Archive/Gallery_archive_8_5.py +0 -446
- pages/Gallery.py +330 -324
- pages/__pycache__/Gallery.cpython-39.pyc +0 -0
- pages/streamlit-1.25.py +13 -13
Archive/Gallery_archive_8_5.py
DELETED
@@ -1,446 +0,0 @@
|
|
1 |
-
import os
|
2 |
-
import requests
|
3 |
-
|
4 |
-
import altair as alt
|
5 |
-
import numpy as np
|
6 |
-
import pandas as pd
|
7 |
-
import streamlit as st
|
8 |
-
|
9 |
-
from bs4 import BeautifulSoup
|
10 |
-
from datasets import load_dataset, Dataset, load_from_disk
|
11 |
-
from huggingface_hub import login
|
12 |
-
from streamlit_extras.switch_page_button import switch_page
|
13 |
-
from sklearn.svm import LinearSVC
|
14 |
-
|
15 |
-
SCORE_NAME_MAPPING = {'clip': 'clip_score', 'rank': 'msq_score', 'pop': 'model_download_count'}
|
16 |
-
|
17 |
-
|
18 |
-
class GalleryApp:
|
19 |
-
def __init__(self, promptBook, images_ds):
|
20 |
-
self.promptBook = promptBook
|
21 |
-
self.images_ds = images_ds
|
22 |
-
|
23 |
-
def gallery_standard(self, items, col_num, info):
|
24 |
-
rows = len(items) // col_num + 1
|
25 |
-
containers = [st.container() for _ in range(rows)]
|
26 |
-
for idx in range(0, len(items), col_num):
|
27 |
-
row_idx = idx // col_num
|
28 |
-
with containers[row_idx]:
|
29 |
-
cols = st.columns(col_num)
|
30 |
-
for j in range(col_num):
|
31 |
-
if idx + j < len(items):
|
32 |
-
with cols[j]:
|
33 |
-
# show image
|
34 |
-
# image = self.images_ds[items.iloc[idx + j]['row_idx'].item()]['image']
|
35 |
-
# image = f"https://modelcofferbucket.s3.us-east-2.amazonaws.com/{items.iloc[idx + j]['image_id']}.png"
|
36 |
-
image = f"https://modelcofferbucket.s3-accelerate.amazonaws.com/{items.iloc[idx + j]['image_id']}.png"
|
37 |
-
st.image(image, use_column_width=True)
|
38 |
-
|
39 |
-
# handel checkbox information
|
40 |
-
prompt_id = items.iloc[idx + j]['prompt_id']
|
41 |
-
modelVersion_id = items.iloc[idx + j]['modelVersion_id']
|
42 |
-
|
43 |
-
check_init = True if modelVersion_id in st.session_state.selected_dict.get(prompt_id, []) else False
|
44 |
-
|
45 |
-
st.write("Position: ", idx + j)
|
46 |
-
|
47 |
-
# show checkbox
|
48 |
-
st.checkbox('Select', key=f'select_{prompt_id}_{modelVersion_id}', value=check_init)
|
49 |
-
|
50 |
-
# show selected info
|
51 |
-
for key in info:
|
52 |
-
st.write(f"**{key}**: {items.iloc[idx + j][key]}")
|
53 |
-
|
54 |
-
def selection_panel(self, items):
|
55 |
-
# temperal function
|
56 |
-
|
57 |
-
selecters = st.columns([1, 4])
|
58 |
-
|
59 |
-
if 'score_weights' not in st.session_state:
|
60 |
-
st.session_state.score_weights = [1.0, 0.8, 0.2, 0.8]
|
61 |
-
|
62 |
-
# select sort type
|
63 |
-
with selecters[0]:
|
64 |
-
sort_type = st.selectbox('Sort by', ['Scores', 'IDs and Names'])
|
65 |
-
if sort_type == 'Scores':
|
66 |
-
sort_by = 'weighted_score_sum'
|
67 |
-
|
68 |
-
# select other options
|
69 |
-
with selecters[1]:
|
70 |
-
if sort_type == 'IDs and Names':
|
71 |
-
sub_selecters = st.columns([3, 1])
|
72 |
-
# select sort by
|
73 |
-
with sub_selecters[0]:
|
74 |
-
sort_by = st.selectbox('Sort by',
|
75 |
-
['model_name', 'model_id', 'modelVersion_name', 'modelVersion_id', 'norm_nsfw'],
|
76 |
-
label_visibility='hidden')
|
77 |
-
|
78 |
-
continue_idx = 1
|
79 |
-
|
80 |
-
else:
|
81 |
-
# add custom weights
|
82 |
-
sub_selecters = st.columns([1, 1, 1, 1])
|
83 |
-
|
84 |
-
with sub_selecters[0]:
|
85 |
-
clip_weight = st.number_input('Clip Score Weight', min_value=-100.0, max_value=100.0, value=st.session_state.score_weights[0], step=0.1, help='the weight for normalized clip score')
|
86 |
-
with sub_selecters[1]:
|
87 |
-
mcos_weight = st.number_input('Dissimilarity Weight', min_value=-100.0, max_value=100.0, value=st.session_state.score_weights[1], step=0.1, help='the weight for m(eam) s(imilarity) q(antile) score for measuring distinctiveness')
|
88 |
-
with sub_selecters[2]:
|
89 |
-
pop_weight = st.number_input('Popularity Weight', min_value=-100.0, max_value=100.0, value=st.session_state.score_weights[2], step=0.1, help='the weight for normalized popularity score')
|
90 |
-
|
91 |
-
items.loc[:, 'weighted_score_sum'] = round(items[f'norm_clip'] * clip_weight + items[f'norm_mcos'] * mcos_weight + items[
|
92 |
-
'norm_pop'] * pop_weight, 4)
|
93 |
-
|
94 |
-
continue_idx = 3
|
95 |
-
|
96 |
-
# save latest weights
|
97 |
-
st.session_state.score_weights[0] = clip_weight
|
98 |
-
st.session_state.score_weights[1] = mcos_weight
|
99 |
-
st.session_state.score_weights[2] = pop_weight
|
100 |
-
|
101 |
-
# select threshold
|
102 |
-
with sub_selecters[continue_idx]:
|
103 |
-
nsfw_threshold = st.number_input('NSFW Score Threshold', min_value=0.0, max_value=1.0, value=st.session_state.score_weights[3], step=0.01, help='Only show models with nsfw score lower than this threshold, set 1.0 to show all images')
|
104 |
-
items = items[items['norm_nsfw'] <= nsfw_threshold].reset_index(drop=True)
|
105 |
-
|
106 |
-
# save latest threshold
|
107 |
-
st.session_state.score_weights[3] = nsfw_threshold
|
108 |
-
|
109 |
-
# draw a distribution histogram
|
110 |
-
if sort_type == 'Scores':
|
111 |
-
try:
|
112 |
-
with st.expander('Show score distribution histogram and select score range'):
|
113 |
-
st.write('**Score distribution histogram**')
|
114 |
-
chart_space = st.container()
|
115 |
-
# st.write('Select the range of scores to show')
|
116 |
-
hist_data = pd.DataFrame(items[sort_by])
|
117 |
-
mini = hist_data[sort_by].min().item()
|
118 |
-
mini = mini//0.1 * 0.1
|
119 |
-
maxi = hist_data[sort_by].max().item()
|
120 |
-
maxi = maxi//0.1 * 0.1 + 0.1
|
121 |
-
st.write('**Select the range of scores to show**')
|
122 |
-
r = st.slider('Select the range of scores to show', min_value=mini, max_value=maxi, value=(mini, maxi), step=0.05, label_visibility='collapsed')
|
123 |
-
with chart_space:
|
124 |
-
st.altair_chart(altair_histogram(hist_data, sort_by, r[0], r[1]), use_container_width=True)
|
125 |
-
# event_dict = altair_component(altair_chart=altair_histogram(hist_data, sort_by))
|
126 |
-
# r = event_dict.get(sort_by)
|
127 |
-
if r:
|
128 |
-
items = items[(items[sort_by] >= r[0]) & (items[sort_by] <= r[1])].reset_index(drop=True)
|
129 |
-
# st.write(r)
|
130 |
-
except:
|
131 |
-
pass
|
132 |
-
|
133 |
-
display_options = st.columns([1, 4])
|
134 |
-
|
135 |
-
with display_options[0]:
|
136 |
-
# select order
|
137 |
-
order = st.selectbox('Order', ['Ascending', 'Descending'], index=1 if sort_type == 'Scores' else 0)
|
138 |
-
if order == 'Ascending':
|
139 |
-
order = True
|
140 |
-
else:
|
141 |
-
order = False
|
142 |
-
|
143 |
-
with display_options[1]:
|
144 |
-
|
145 |
-
# select info to show
|
146 |
-
info = st.multiselect('Show Info',
|
147 |
-
['model_name', 'model_id', 'modelVersion_name', 'modelVersion_id',
|
148 |
-
'weighted_score_sum', 'model_download_count', 'clip_score', 'mcos_score',
|
149 |
-
'nsfw_score', 'norm_nsfw'],
|
150 |
-
default=sort_by)
|
151 |
-
|
152 |
-
# apply sorting to dataframe
|
153 |
-
items = items.sort_values(by=[sort_by], ascending=order).reset_index(drop=True)
|
154 |
-
|
155 |
-
# select number of columns
|
156 |
-
col_num = st.slider('Number of columns', min_value=1, max_value=9, value=4, step=1, key='col_num')
|
157 |
-
|
158 |
-
return items, info, col_num
|
159 |
-
|
160 |
-
def sidebar(self):
|
161 |
-
with st.sidebar:
|
162 |
-
prompt_tags = self.promptBook['tag'].unique()
|
163 |
-
# sort tags by alphabetical order
|
164 |
-
prompt_tags = np.sort(prompt_tags)[::-1]
|
165 |
-
|
166 |
-
tag = st.selectbox('Select a tag', prompt_tags)
|
167 |
-
|
168 |
-
items = self.promptBook[self.promptBook['tag'] == tag].reset_index(drop=True)
|
169 |
-
|
170 |
-
prompts = np.sort(items['prompt'].unique())[::-1]
|
171 |
-
|
172 |
-
selected_prompt = st.selectbox('Select prompt', prompts)
|
173 |
-
|
174 |
-
items = items[items['prompt'] == selected_prompt].reset_index(drop=True)
|
175 |
-
prompt_id = items['prompt_id'].unique()[0]
|
176 |
-
note = items['note'].unique()[0]
|
177 |
-
|
178 |
-
# show source
|
179 |
-
if isinstance(note, str):
|
180 |
-
if note.isdigit():
|
181 |
-
st.caption(f"`Source: civitai`")
|
182 |
-
else:
|
183 |
-
st.caption(f"`Source: {note}`")
|
184 |
-
else:
|
185 |
-
st.caption("`Source: Parti-prompts`")
|
186 |
-
|
187 |
-
# show image metadata
|
188 |
-
image_metadatas = ['prompt_id', 'prompt', 'negativePrompt', 'sampler', 'cfgScale', 'size', 'seed']
|
189 |
-
for key in image_metadatas:
|
190 |
-
label = ' '.join(key.split('_')).capitalize()
|
191 |
-
st.write(f"**{label}**")
|
192 |
-
if items[key][0] == ' ':
|
193 |
-
st.write('`None`')
|
194 |
-
else:
|
195 |
-
st.caption(f"{items[key][0]}")
|
196 |
-
|
197 |
-
# for note as civitai image id, add civitai reference
|
198 |
-
if isinstance(note, str) and note.isdigit():
|
199 |
-
try:
|
200 |
-
st.write(f'**[Civitai Reference](https://civitai.com/images/{note})**')
|
201 |
-
res = requests.get(f'https://civitai.com/images/{note}')
|
202 |
-
# st.write(res.text)
|
203 |
-
soup = BeautifulSoup(res.text, 'html.parser')
|
204 |
-
image_section = soup.find('div', {'class': 'mantine-12rlksp'})
|
205 |
-
image_url = image_section.find('img')['src']
|
206 |
-
st.image(image_url, use_column_width=True)
|
207 |
-
except:
|
208 |
-
pass
|
209 |
-
|
210 |
-
return prompt_tags, tag, prompt_id, items
|
211 |
-
|
212 |
-
def app(self):
|
213 |
-
st.title('Model Visualization and Retrieval')
|
214 |
-
st.write('This is a gallery of images generated by the models')
|
215 |
-
|
216 |
-
prompt_tags, tag, prompt_id, items = self.sidebar()
|
217 |
-
|
218 |
-
# add safety check for some prompts
|
219 |
-
safety_check = True
|
220 |
-
unsafe_prompts = {}
|
221 |
-
# initialize unsafe prompts
|
222 |
-
for prompt_tag in prompt_tags:
|
223 |
-
unsafe_prompts[prompt_tag] = []
|
224 |
-
# manually add unsafe prompts
|
225 |
-
unsafe_prompts['world knowledge'] = [83]
|
226 |
-
# unsafe_prompts['art'] = [23]
|
227 |
-
unsafe_prompts['abstract'] = [1, 3]
|
228 |
-
# unsafe_prompts['food'] = [34]
|
229 |
-
|
230 |
-
if int(prompt_id.item()) in unsafe_prompts[tag]:
|
231 |
-
st.warning('This prompt may contain unsafe content. They might be offensive, depressing, or sexual.')
|
232 |
-
safety_check = st.checkbox('I understand that this prompt may contain unsafe content. Show these images anyway.', key=f'{prompt_id}')
|
233 |
-
|
234 |
-
if safety_check:
|
235 |
-
items, info, col_num = self.selection_panel(items)
|
236 |
-
|
237 |
-
if 'selected_dict' in st.session_state:
|
238 |
-
st.write('checked: ', str(st.session_state.selected_dict.get(prompt_id, [])))
|
239 |
-
dynamic_weight_options = ['Grid Search', 'SVM', 'Greedy']
|
240 |
-
dynamic_weight_panel = st.columns(len(dynamic_weight_options))
|
241 |
-
|
242 |
-
if len(st.session_state.selected_dict.get(prompt_id, [])) > 0:
|
243 |
-
btn_disable = False
|
244 |
-
else:
|
245 |
-
btn_disable = True
|
246 |
-
|
247 |
-
for i in range(len(dynamic_weight_options)):
|
248 |
-
method = dynamic_weight_options[i]
|
249 |
-
with dynamic_weight_panel[i]:
|
250 |
-
btn = st.button(method, use_container_width=True, disabled=btn_disable, on_click=self.dynamic_weight, args=(prompt_id, items, method))
|
251 |
-
|
252 |
-
with st.form(key=f'{prompt_id}'):
|
253 |
-
# buttons = st.columns([1, 1, 1])
|
254 |
-
buttons_space = st.columns([1, 1, 1, 1])
|
255 |
-
gallery_space = st.empty()
|
256 |
-
|
257 |
-
with buttons_space[0]:
|
258 |
-
continue_btn = st.form_submit_button('Confirm Selection', use_container_width=True, type='primary')
|
259 |
-
if continue_btn:
|
260 |
-
self.submit_actions('Continue', prompt_id)
|
261 |
-
|
262 |
-
with buttons_space[1]:
|
263 |
-
select_btn = st.form_submit_button('Select All', use_container_width=True)
|
264 |
-
if select_btn:
|
265 |
-
self.submit_actions('Select', prompt_id)
|
266 |
-
|
267 |
-
with buttons_space[2]:
|
268 |
-
deselect_btn = st.form_submit_button('Deselect All', use_container_width=True)
|
269 |
-
if deselect_btn:
|
270 |
-
self.submit_actions('Deselect', prompt_id)
|
271 |
-
|
272 |
-
with buttons_space[3]:
|
273 |
-
refresh_btn = st.form_submit_button('Refresh', on_click=gallery_space.empty, use_container_width=True)
|
274 |
-
|
275 |
-
with gallery_space.container():
|
276 |
-
with st.spinner('Loading images...'):
|
277 |
-
self.gallery_standard(items, col_num, info)
|
278 |
-
|
279 |
-
def submit_actions(self, status, prompt_id):
|
280 |
-
if status == 'Select':
|
281 |
-
modelVersions = self.promptBook[self.promptBook['prompt_id'] == prompt_id]['modelVersion_id'].unique()
|
282 |
-
st.session_state.selected_dict[prompt_id] = modelVersions.tolist()
|
283 |
-
print(st.session_state.selected_dict, 'select')
|
284 |
-
st.experimental_rerun()
|
285 |
-
elif status == 'Deselect':
|
286 |
-
st.session_state.selected_dict[prompt_id] = []
|
287 |
-
print(st.session_state.selected_dict, 'deselect')
|
288 |
-
st.experimental_rerun()
|
289 |
-
# self.promptBook.loc[self.promptBook['prompt_id'] == prompt_id, 'checked'] = False
|
290 |
-
elif status == 'Continue':
|
291 |
-
st.session_state.selected_dict[prompt_id] = []
|
292 |
-
for key in st.session_state:
|
293 |
-
keys = key.split('_')
|
294 |
-
if keys[0] == 'select' and keys[1] == str(prompt_id):
|
295 |
-
if st.session_state[key]:
|
296 |
-
st.session_state.selected_dict[prompt_id].append(int(keys[2]))
|
297 |
-
# switch_page("ranking")
|
298 |
-
print(st.session_state.selected_dict, 'continue')
|
299 |
-
st.experimental_rerun()
|
300 |
-
|
301 |
-
def dynamic_weight(self, prompt_id, items, method='Grid Search'):
|
302 |
-
selected = items[
|
303 |
-
items['modelVersion_id'].isin(st.session_state.selected_dict[prompt_id])].reset_index(drop=True)
|
304 |
-
optimal_weight = [0, 0, 0]
|
305 |
-
|
306 |
-
if method == 'Grid Search':
|
307 |
-
# grid search method
|
308 |
-
top_ranking = len(items) * len(selected)
|
309 |
-
|
310 |
-
for clip_weight in np.arange(-1, 1, 0.1):
|
311 |
-
for mcos_weight in np.arange(-1, 1, 0.1):
|
312 |
-
for pop_weight in np.arange(-1, 1, 0.1):
|
313 |
-
|
314 |
-
weight_all = clip_weight*items[f'norm_clip'] + mcos_weight*items[f'norm_mcos'] + pop_weight*items['norm_pop']
|
315 |
-
weight_all_sorted = weight_all.sort_values(ascending=False).reset_index(drop=True)
|
316 |
-
# print('weight_all_sorted:', weight_all_sorted)
|
317 |
-
weight_selected = clip_weight*selected[f'norm_clip'] + mcos_weight*selected[f'norm_mcos'] + pop_weight*selected['norm_pop']
|
318 |
-
|
319 |
-
# get the index of values of weight_selected in weight_all_sorted
|
320 |
-
rankings = []
|
321 |
-
for weight in weight_selected:
|
322 |
-
rankings.append(weight_all_sorted.index[weight_all_sorted == weight].tolist()[0])
|
323 |
-
if sum(rankings) <= top_ranking:
|
324 |
-
top_ranking = sum(rankings)
|
325 |
-
print('current top ranking:', top_ranking, rankings)
|
326 |
-
optimal_weight = [clip_weight, mcos_weight, pop_weight]
|
327 |
-
print('optimal weight:', optimal_weight)
|
328 |
-
|
329 |
-
elif method == 'SVM':
|
330 |
-
# svm method
|
331 |
-
print('start svm method')
|
332 |
-
# get residual dataframe that contains models not selected
|
333 |
-
residual = items[~items['modelVersion_id'].isin(selected['modelVersion_id'])].reset_index(drop=True)
|
334 |
-
residual = residual[['norm_clip_crop', 'norm_mcos_crop', 'norm_pop']]
|
335 |
-
residual = residual.to_numpy()
|
336 |
-
selected = selected[['norm_clip_crop', 'norm_mcos_crop', 'norm_pop']]
|
337 |
-
selected = selected.to_numpy()
|
338 |
-
|
339 |
-
y = np.concatenate((np.full((len(selected), 1), -1), np.full((len(residual), 1), 1)), axis=0).ravel()
|
340 |
-
X = np.concatenate((selected, residual), axis=0)
|
341 |
-
|
342 |
-
# fit svm model, and get parameters for the hyperplane
|
343 |
-
clf = LinearSVC(random_state=0, C=1.0, fit_intercept=False, dual='auto')
|
344 |
-
clf.fit(X, y)
|
345 |
-
optimal_weight = clf.coef_[0].tolist()
|
346 |
-
print('optimal weight:', optimal_weight)
|
347 |
-
pass
|
348 |
-
|
349 |
-
elif method == 'Greedy':
|
350 |
-
for idx in selected.index:
|
351 |
-
# find which score is the highest, clip, mcos, or pop
|
352 |
-
clip_score = selected.loc[idx, 'norm_clip_crop']
|
353 |
-
mcos_score = selected.loc[idx, 'norm_mcos_crop']
|
354 |
-
pop_score = selected.loc[idx, 'norm_pop']
|
355 |
-
if clip_score >= mcos_score and clip_score >= pop_score:
|
356 |
-
optimal_weight[0] += 1
|
357 |
-
elif mcos_score >= clip_score and mcos_score >= pop_score:
|
358 |
-
optimal_weight[1] += 1
|
359 |
-
elif pop_score >= clip_score and pop_score >= mcos_score:
|
360 |
-
optimal_weight[2] += 1
|
361 |
-
|
362 |
-
# normalize optimal_weight
|
363 |
-
optimal_weight = [round(weight/len(selected), 2) for weight in optimal_weight]
|
364 |
-
print('optimal weight:', optimal_weight)
|
365 |
-
|
366 |
-
st.session_state.score_weights[0: 3] = optimal_weight
|
367 |
-
|
368 |
-
|
369 |
-
# hist_data = pd.DataFrame(np.random.normal(42, 10, (200, 1)), columns=["x"])
|
370 |
-
@st.cache_resource
|
371 |
-
def altair_histogram(hist_data, sort_by, mini, maxi):
|
372 |
-
brushed = alt.selection_interval(encodings=['x'], name="brushed")
|
373 |
-
|
374 |
-
chart = (
|
375 |
-
alt.Chart(hist_data)
|
376 |
-
.mark_bar(opacity=0.7, cornerRadius=2)
|
377 |
-
.encode(alt.X(f"{sort_by}:Q", bin=alt.Bin(maxbins=25)), y="count()")
|
378 |
-
# .add_selection(brushed)
|
379 |
-
# .properties(width=800, height=300)
|
380 |
-
)
|
381 |
-
|
382 |
-
# Create a transparent rectangle for highlighting the range
|
383 |
-
highlight = (
|
384 |
-
alt.Chart(pd.DataFrame({'x1': [mini], 'x2': [maxi]}))
|
385 |
-
.mark_rect(opacity=0.3)
|
386 |
-
.encode(x='x1', x2='x2')
|
387 |
-
# .properties(width=800, height=300)
|
388 |
-
)
|
389 |
-
|
390 |
-
# Layer the chart and the highlight rectangle
|
391 |
-
layered_chart = alt.layer(chart, highlight)
|
392 |
-
|
393 |
-
return layered_chart
|
394 |
-
|
395 |
-
|
396 |
-
@st.cache_data
|
397 |
-
def load_hf_dataset():
|
398 |
-
# login to huggingface
|
399 |
-
login(token=os.environ.get("HF_TOKEN"))
|
400 |
-
|
401 |
-
# load from huggingface
|
402 |
-
roster = pd.DataFrame(load_dataset('NYUSHPRP/ModelCofferRoster', split='train'))
|
403 |
-
promptBook = pd.DataFrame(load_dataset('NYUSHPRP/ModelCofferMetadata', split='train'))
|
404 |
-
# images_ds = load_from_disk(os.path.join(os.getcwd(), 'data', 'promptbook'))
|
405 |
-
images_ds = None # set to None for now since we use s3 bucket to store images
|
406 |
-
|
407 |
-
# process dataset
|
408 |
-
roster = roster[['model_id', 'model_name', 'modelVersion_id', 'modelVersion_name',
|
409 |
-
'model_download_count']].drop_duplicates().reset_index(drop=True)
|
410 |
-
|
411 |
-
# add 'custom_score_weights' column to promptBook if not exist
|
412 |
-
if 'weighted_score_sum' not in promptBook.columns:
|
413 |
-
promptBook.loc[:, 'weighted_score_sum'] = 0
|
414 |
-
|
415 |
-
# merge roster and promptbook
|
416 |
-
promptBook = promptBook.merge(roster[['model_id', 'model_name', 'modelVersion_id', 'modelVersion_name', 'model_download_count']],
|
417 |
-
on=['model_id', 'modelVersion_id'], how='left')
|
418 |
-
|
419 |
-
# add column to record current row index
|
420 |
-
promptBook.loc[:, 'row_idx'] = promptBook.index
|
421 |
-
|
422 |
-
return roster, promptBook, images_ds
|
423 |
-
|
424 |
-
|
425 |
-
if __name__ == "__main__":
|
426 |
-
st.set_page_config(page_title="Model Coffer Gallery", page_icon="🖼️", layout="wide")
|
427 |
-
|
428 |
-
# remove ranking in the session state if it is created in Ranking.py
|
429 |
-
st.session_state.pop('ranking', None)
|
430 |
-
|
431 |
-
if 'user_id' not in st.session_state:
|
432 |
-
st.warning('Please log in first.')
|
433 |
-
home_btn = st.button('Go to Home Page')
|
434 |
-
if home_btn:
|
435 |
-
switch_page("home")
|
436 |
-
else:
|
437 |
-
st.write('You have already logged in as ' + st.session_state.user_id[0])
|
438 |
-
roster, promptBook, images_ds = load_hf_dataset()
|
439 |
-
# print(promptBook.columns)
|
440 |
-
|
441 |
-
# initialize selected_dict
|
442 |
-
if 'selected_dict' not in st.session_state:
|
443 |
-
st.session_state['selected_dict'] = {}
|
444 |
-
|
445 |
-
app = GalleryApp(promptBook=promptBook, images_ds=images_ds)
|
446 |
-
app.app()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
pages/Gallery.py
CHANGED
@@ -14,353 +14,358 @@ from sklearn.svm import LinearSVC
|
|
14 |
|
15 |
SCORE_NAME_MAPPING = {'clip': 'clip_score', 'rank': 'msq_score', 'pop': 'model_download_count'}
|
16 |
|
17 |
-
def gallery_standard(items, col_num, info):
|
18 |
-
rows = len(items) // col_num + 1
|
19 |
-
containers = [st.container() for _ in range(rows)]
|
20 |
-
for idx in range(0, len(items), col_num):
|
21 |
-
row_idx = idx // col_num
|
22 |
-
with containers[row_idx]:
|
23 |
-
cols = st.columns(col_num)
|
24 |
-
for j in range(col_num):
|
25 |
-
if idx + j < len(items):
|
26 |
-
with cols[j]:
|
27 |
-
# show image
|
28 |
-
# image = self.images_ds[items.iloc[idx + j]['row_idx'].item()]['image']
|
29 |
-
# image = f"https://modelcofferbucket.s3.us-east-2.amazonaws.com/{items.iloc[idx + j]['image_id']}.png"
|
30 |
-
image = f"https://modelcofferbucket.s3-accelerate.amazonaws.com/{items.iloc[idx + j]['image_id']}.png"
|
31 |
-
st.image(image, use_column_width=True)
|
32 |
-
|
33 |
-
# handel checkbox information
|
34 |
-
prompt_id = items.iloc[idx + j]['prompt_id']
|
35 |
-
modelVersion_id = items.iloc[idx + j]['modelVersion_id']
|
36 |
-
|
37 |
-
check_init = True if modelVersion_id in st.session_state.selected_dict.get(prompt_id, []) else False
|
38 |
-
|
39 |
-
st.write("Position: ", idx + j)
|
40 |
-
|
41 |
-
# show checkbox
|
42 |
-
st.checkbox('Select', key=f'select_{prompt_id}_{modelVersion_id}', value=check_init)
|
43 |
-
|
44 |
-
# show selected info
|
45 |
-
for key in info:
|
46 |
-
st.write(f"**{key}**: {items.iloc[idx + j][key]}")
|
47 |
-
|
48 |
-
def selection_panel(items):
|
49 |
-
# temperal function
|
50 |
-
|
51 |
-
selecters = st.columns([1, 4])
|
52 |
-
|
53 |
-
if 'score_weights' not in st.session_state:
|
54 |
-
st.session_state.score_weights = [1.0, 0.8, 0.2, 0.8]
|
55 |
-
|
56 |
-
# select sort type
|
57 |
-
with selecters[0]:
|
58 |
-
sort_type = st.selectbox('Sort by', ['Scores', 'IDs and Names'])
|
59 |
-
if sort_type == 'Scores':
|
60 |
-
sort_by = 'weighted_score_sum'
|
61 |
-
|
62 |
-
# select other options
|
63 |
-
with selecters[1]:
|
64 |
-
if sort_type == 'IDs and Names':
|
65 |
-
sub_selecters = st.columns([3, 1])
|
66 |
-
# select sort by
|
67 |
-
with sub_selecters[0]:
|
68 |
-
sort_by = st.selectbox('Sort by',
|
69 |
-
['model_name', 'model_id', 'modelVersion_name', 'modelVersion_id', 'norm_nsfw'],
|
70 |
-
label_visibility='hidden')
|
71 |
-
|
72 |
-
continue_idx = 1
|
73 |
-
|
74 |
-
else:
|
75 |
-
# add custom weights
|
76 |
-
sub_selecters = st.columns([1, 1, 1, 1])
|
77 |
-
|
78 |
-
with sub_selecters[0]:
|
79 |
-
clip_weight = st.number_input('Clip Score Weight', min_value=-100.0, max_value=100.0, value=st.session_state.score_weights[0], step=0.1, help='the weight for normalized clip score')
|
80 |
-
with sub_selecters[1]:
|
81 |
-
mcos_weight = st.number_input('Dissimilarity Weight', min_value=-100.0, max_value=100.0, value=st.session_state.score_weights[1], step=0.1, help='the weight for m(eam) s(imilarity) q(antile) score for measuring distinctiveness')
|
82 |
-
with sub_selecters[2]:
|
83 |
-
pop_weight = st.number_input('Popularity Weight', min_value=-100.0, max_value=100.0, value=st.session_state.score_weights[2], step=0.1, help='the weight for normalized popularity score')
|
84 |
-
|
85 |
-
items.loc[:, 'weighted_score_sum'] = round(items[f'norm_clip'] * clip_weight + items[f'norm_mcos'] * mcos_weight + items[
|
86 |
-
'norm_pop'] * pop_weight, 4)
|
87 |
-
|
88 |
-
continue_idx = 3
|
89 |
-
|
90 |
-
# save latest weights
|
91 |
-
st.session_state.score_weights[0] = clip_weight
|
92 |
-
st.session_state.score_weights[1] = mcos_weight
|
93 |
-
st.session_state.score_weights[2] = pop_weight
|
94 |
-
|
95 |
-
# select threshold
|
96 |
-
with sub_selecters[continue_idx]:
|
97 |
-
nsfw_threshold = st.number_input('NSFW Score Threshold', min_value=0.0, max_value=1.0, value=st.session_state.score_weights[3], step=0.01, help='Only show models with nsfw score lower than this threshold, set 1.0 to show all images')
|
98 |
-
items = items[items['norm_nsfw'] <= nsfw_threshold].reset_index(drop=True)
|
99 |
-
|
100 |
-
# save latest threshold
|
101 |
-
st.session_state.score_weights[3] = nsfw_threshold
|
102 |
-
|
103 |
-
# draw a distribution histogram
|
104 |
-
if sort_type == 'Scores':
|
105 |
-
try:
|
106 |
-
with st.expander('Show score distribution histogram and select score range'):
|
107 |
-
st.write('**Score distribution histogram**')
|
108 |
-
chart_space = st.container()
|
109 |
-
# st.write('Select the range of scores to show')
|
110 |
-
hist_data = pd.DataFrame(items[sort_by])
|
111 |
-
mini = hist_data[sort_by].min().item()
|
112 |
-
mini = mini//0.1 * 0.1
|
113 |
-
maxi = hist_data[sort_by].max().item()
|
114 |
-
maxi = maxi//0.1 * 0.1 + 0.1
|
115 |
-
st.write('**Select the range of scores to show**')
|
116 |
-
r = st.slider('Select the range of scores to show', min_value=mini, max_value=maxi, value=(mini, maxi), step=0.05, label_visibility='collapsed')
|
117 |
-
with chart_space:
|
118 |
-
st.altair_chart(altair_histogram(hist_data, sort_by, r[0], r[1]), use_container_width=True)
|
119 |
-
# event_dict = altair_component(altair_chart=altair_histogram(hist_data, sort_by))
|
120 |
-
# r = event_dict.get(sort_by)
|
121 |
-
if r:
|
122 |
-
items = items[(items[sort_by] >= r[0]) & (items[sort_by] <= r[1])].reset_index(drop=True)
|
123 |
-
# st.write(r)
|
124 |
-
except:
|
125 |
-
pass
|
126 |
|
127 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
128 |
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
if order == 'Ascending':
|
133 |
-
order = True
|
134 |
-
else:
|
135 |
-
order = False
|
136 |
|
137 |
-
|
|
|
|
|
|
|
|
|
|
|
138 |
|
139 |
-
|
140 |
-
|
141 |
-
['model_name', 'model_id', 'modelVersion_name', 'modelVersion_id',
|
142 |
-
'weighted_score_sum', 'model_download_count', 'clip_score', 'mcos_score',
|
143 |
-
'nsfw_score', 'norm_nsfw'],
|
144 |
-
default=sort_by)
|
145 |
|
146 |
-
|
147 |
-
items = items.sort_values(by=[sort_by], ascending=order).reset_index(drop=True)
|
148 |
|
149 |
-
|
150 |
-
|
|
|
|
|
151 |
|
152 |
-
|
|
|
|
|
|
|
153 |
|
154 |
-
|
155 |
-
|
156 |
-
prompt_tags = promptBook['tag'].unique()
|
157 |
-
# sort tags by alphabetical order
|
158 |
-
prompt_tags = np.sort(prompt_tags)[::-1]
|
159 |
|
160 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
161 |
|
162 |
-
|
163 |
|
164 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
165 |
|
166 |
-
|
167 |
|
168 |
-
|
169 |
-
|
170 |
-
|
|
|
|
|
|
|
171 |
|
172 |
-
#
|
173 |
-
|
174 |
-
if note.isdigit():
|
175 |
-
st.caption(f"`Source: civitai`")
|
176 |
-
else:
|
177 |
-
st.caption(f"`Source: {note}`")
|
178 |
-
else:
|
179 |
-
st.caption("`Source: Parti-prompts`")
|
180 |
-
|
181 |
-
# show image metadata
|
182 |
-
image_metadatas = ['prompt_id', 'prompt', 'negativePrompt', 'sampler', 'cfgScale', 'size', 'seed']
|
183 |
-
for key in image_metadatas:
|
184 |
-
label = ' '.join(key.split('_')).capitalize()
|
185 |
-
st.write(f"**{label}**")
|
186 |
-
if items[key][0] == ' ':
|
187 |
-
st.write('`None`')
|
188 |
-
else:
|
189 |
-
st.caption(f"{items[key][0]}")
|
190 |
|
191 |
-
#
|
192 |
-
|
193 |
-
try:
|
194 |
-
st.write(f'**[Civitai Reference](https://civitai.com/images/{note})**')
|
195 |
-
res = requests.get(f'https://civitai.com/images/{note}')
|
196 |
-
# st.write(res.text)
|
197 |
-
soup = BeautifulSoup(res.text, 'html.parser')
|
198 |
-
image_section = soup.find('div', {'class': 'mantine-12rlksp'})
|
199 |
-
image_url = image_section.find('img')['src']
|
200 |
-
st.image(image_url, use_column_width=True)
|
201 |
-
except:
|
202 |
-
pass
|
203 |
|
204 |
-
|
205 |
|
206 |
-
def
|
207 |
-
|
208 |
-
|
|
|
|
|
209 |
|
210 |
-
|
211 |
|
212 |
-
|
213 |
-
safety_check = True
|
214 |
-
unsafe_prompts = {}
|
215 |
-
# initialize unsafe prompts
|
216 |
-
for prompt_tag in prompt_tags:
|
217 |
-
unsafe_prompts[prompt_tag] = []
|
218 |
-
# manually add unsafe prompts
|
219 |
-
unsafe_prompts['world knowledge'] = [83]
|
220 |
-
# unsafe_prompts['art'] = [23]
|
221 |
-
unsafe_prompts['abstract'] = [1, 3]
|
222 |
-
# unsafe_prompts['food'] = [34]
|
223 |
|
224 |
-
|
225 |
-
st.warning('This prompt may contain unsafe content. They might be offensive, depressing, or sexual.')
|
226 |
-
safety_check = st.checkbox('I understand that this prompt may contain unsafe content. Show these images anyway.', key=f'{prompt_id}')
|
227 |
|
228 |
-
|
229 |
-
items, info, col_num = selection_panel(items)
|
230 |
|
231 |
-
|
232 |
-
|
233 |
-
|
234 |
-
dynamic_weight_panel = st.columns(len(dynamic_weight_options))
|
235 |
|
236 |
-
|
237 |
-
|
|
|
|
|
|
|
|
|
238 |
else:
|
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 |
-
if
|
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 |
-
optimal_weight = [
|
361 |
-
|
362 |
-
|
363 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
364 |
|
365 |
|
366 |
# hist_data = pd.DataFrame(np.random.normal(42, 10, (200, 1)), columns=["x"])
|
@@ -439,4 +444,5 @@ if __name__ == "__main__":
|
|
439 |
if 'selected_dict' not in st.session_state:
|
440 |
st.session_state['selected_dict'] = {}
|
441 |
|
442 |
-
app(promptBook, images_ds)
|
|
|
|
14 |
|
15 |
SCORE_NAME_MAPPING = {'clip': 'clip_score', 'rank': 'msq_score', 'pop': 'model_download_count'}
|
16 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
|
18 |
+
class GalleryApp:
|
19 |
+
def __init__(self, promptBook, images_ds):
|
20 |
+
self.promptBook = promptBook
|
21 |
+
self.images_ds = images_ds
|
22 |
+
|
23 |
+
def gallery_standard(self, items, col_num, info):
|
24 |
+
rows = len(items) // col_num + 1
|
25 |
+
containers = [st.container() for _ in range(rows)]
|
26 |
+
for idx in range(0, len(items), col_num):
|
27 |
+
row_idx = idx // col_num
|
28 |
+
with containers[row_idx]:
|
29 |
+
cols = st.columns(col_num)
|
30 |
+
for j in range(col_num):
|
31 |
+
if idx + j < len(items):
|
32 |
+
with cols[j]:
|
33 |
+
# show image
|
34 |
+
# image = self.images_ds[items.iloc[idx + j]['row_idx'].item()]['image']
|
35 |
+
# image = f"https://modelcofferbucket.s3.us-east-2.amazonaws.com/{items.iloc[idx + j]['image_id']}.png"
|
36 |
+
image = f"https://modelcofferbucket.s3-accelerate.amazonaws.com/{items.iloc[idx + j]['image_id']}.png"
|
37 |
+
st.image(image, use_column_width=True)
|
38 |
+
|
39 |
+
# handel checkbox information
|
40 |
+
prompt_id = items.iloc[idx + j]['prompt_id']
|
41 |
+
modelVersion_id = items.iloc[idx + j]['modelVersion_id']
|
42 |
+
|
43 |
+
check_init = True if modelVersion_id in st.session_state.selected_dict.get(prompt_id, []) else False
|
44 |
+
|
45 |
+
st.write("Position: ", idx + j)
|
46 |
+
|
47 |
+
# show checkbox
|
48 |
+
st.checkbox('Select', key=f'select_{prompt_id}_{modelVersion_id}', value=check_init)
|
49 |
+
|
50 |
+
# show selected info
|
51 |
+
for key in info:
|
52 |
+
st.write(f"**{key}**: {items.iloc[idx + j][key]}")
|
53 |
+
|
54 |
+
def selection_panel(self, items):
|
55 |
+
# temperal function
|
56 |
+
|
57 |
+
selecters = st.columns([1, 4])
|
58 |
+
|
59 |
+
if 'score_weights' not in st.session_state:
|
60 |
+
st.session_state.score_weights = [1.0, 0.8, 0.2, 0.8]
|
61 |
+
|
62 |
+
# select sort type
|
63 |
+
with selecters[0]:
|
64 |
+
sort_type = st.selectbox('Sort by', ['Scores', 'IDs and Names'])
|
65 |
+
if sort_type == 'Scores':
|
66 |
+
sort_by = 'weighted_score_sum'
|
67 |
+
|
68 |
+
# select other options
|
69 |
+
with selecters[1]:
|
70 |
+
if sort_type == 'IDs and Names':
|
71 |
+
sub_selecters = st.columns([3, 1])
|
72 |
+
# select sort by
|
73 |
+
with sub_selecters[0]:
|
74 |
+
sort_by = st.selectbox('Sort by',
|
75 |
+
['model_name', 'model_id', 'modelVersion_name', 'modelVersion_id', 'norm_nsfw'],
|
76 |
+
label_visibility='hidden')
|
77 |
+
|
78 |
+
continue_idx = 1
|
79 |
|
80 |
+
else:
|
81 |
+
# add custom weights
|
82 |
+
sub_selecters = st.columns([1, 1, 1, 1])
|
|
|
|
|
|
|
|
|
83 |
|
84 |
+
with sub_selecters[0]:
|
85 |
+
clip_weight = st.number_input('Clip Score Weight', min_value=-100.0, max_value=100.0, value=st.session_state.score_weights[0], step=0.1, help='the weight for normalized clip score')
|
86 |
+
with sub_selecters[1]:
|
87 |
+
mcos_weight = st.number_input('Dissimilarity Weight', min_value=-100.0, max_value=100.0, value=st.session_state.score_weights[1], step=0.1, help='the weight for m(eam) s(imilarity) q(antile) score for measuring distinctiveness')
|
88 |
+
with sub_selecters[2]:
|
89 |
+
pop_weight = st.number_input('Popularity Weight', min_value=-100.0, max_value=100.0, value=st.session_state.score_weights[2], step=0.1, help='the weight for normalized popularity score')
|
90 |
|
91 |
+
items.loc[:, 'weighted_score_sum'] = round(items[f'norm_clip'] * clip_weight + items[f'norm_mcos'] * mcos_weight + items[
|
92 |
+
'norm_pop'] * pop_weight, 4)
|
|
|
|
|
|
|
|
|
93 |
|
94 |
+
continue_idx = 3
|
|
|
95 |
|
96 |
+
# save latest weights
|
97 |
+
st.session_state.score_weights[0] = clip_weight
|
98 |
+
st.session_state.score_weights[1] = mcos_weight
|
99 |
+
st.session_state.score_weights[2] = pop_weight
|
100 |
|
101 |
+
# select threshold
|
102 |
+
with sub_selecters[continue_idx]:
|
103 |
+
nsfw_threshold = st.number_input('NSFW Score Threshold', min_value=0.0, max_value=1.0, value=st.session_state.score_weights[3], step=0.01, help='Only show models with nsfw score lower than this threshold, set 1.0 to show all images')
|
104 |
+
items = items[items['norm_nsfw'] <= nsfw_threshold].reset_index(drop=True)
|
105 |
|
106 |
+
# save latest threshold
|
107 |
+
st.session_state.score_weights[3] = nsfw_threshold
|
|
|
|
|
|
|
108 |
|
109 |
+
# draw a distribution histogram
|
110 |
+
if sort_type == 'Scores':
|
111 |
+
try:
|
112 |
+
with st.expander('Show score distribution histogram and select score range'):
|
113 |
+
st.write('**Score distribution histogram**')
|
114 |
+
chart_space = st.container()
|
115 |
+
# st.write('Select the range of scores to show')
|
116 |
+
hist_data = pd.DataFrame(items[sort_by])
|
117 |
+
mini = hist_data[sort_by].min().item()
|
118 |
+
mini = mini//0.1 * 0.1
|
119 |
+
maxi = hist_data[sort_by].max().item()
|
120 |
+
maxi = maxi//0.1 * 0.1 + 0.1
|
121 |
+
st.write('**Select the range of scores to show**')
|
122 |
+
r = st.slider('Select the range of scores to show', min_value=mini, max_value=maxi, value=(mini, maxi), step=0.05, label_visibility='collapsed')
|
123 |
+
with chart_space:
|
124 |
+
st.altair_chart(altair_histogram(hist_data, sort_by, r[0], r[1]), use_container_width=True)
|
125 |
+
# event_dict = altair_component(altair_chart=altair_histogram(hist_data, sort_by))
|
126 |
+
# r = event_dict.get(sort_by)
|
127 |
+
if r:
|
128 |
+
items = items[(items[sort_by] >= r[0]) & (items[sort_by] <= r[1])].reset_index(drop=True)
|
129 |
+
# st.write(r)
|
130 |
+
except:
|
131 |
+
pass
|
132 |
|
133 |
+
display_options = st.columns([1, 4])
|
134 |
|
135 |
+
with display_options[0]:
|
136 |
+
# select order
|
137 |
+
order = st.selectbox('Order', ['Ascending', 'Descending'], index=1 if sort_type == 'Scores' else 0)
|
138 |
+
if order == 'Ascending':
|
139 |
+
order = True
|
140 |
+
else:
|
141 |
+
order = False
|
142 |
|
143 |
+
with display_options[1]:
|
144 |
|
145 |
+
# select info to show
|
146 |
+
info = st.multiselect('Show Info',
|
147 |
+
['model_name', 'model_id', 'modelVersion_name', 'modelVersion_id',
|
148 |
+
'weighted_score_sum', 'model_download_count', 'clip_score', 'mcos_score',
|
149 |
+
'nsfw_score', 'norm_nsfw'],
|
150 |
+
default=sort_by)
|
151 |
|
152 |
+
# apply sorting to dataframe
|
153 |
+
items = items.sort_values(by=[sort_by], ascending=order).reset_index(drop=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
154 |
|
155 |
+
# select number of columns
|
156 |
+
col_num = st.slider('Number of columns', min_value=1, max_value=9, value=4, step=1, key='col_num')
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
157 |
|
158 |
+
return items, info, col_num
|
159 |
|
160 |
+
def sidebar(self):
|
161 |
+
with st.sidebar:
|
162 |
+
prompt_tags = self.promptBook['tag'].unique()
|
163 |
+
# sort tags by alphabetical order
|
164 |
+
prompt_tags = np.sort(prompt_tags)[::-1]
|
165 |
|
166 |
+
tag = st.selectbox('Select a tag', prompt_tags)
|
167 |
|
168 |
+
items = self.promptBook[self.promptBook['tag'] == tag].reset_index(drop=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
169 |
|
170 |
+
prompts = np.sort(items['prompt'].unique())[::-1]
|
|
|
|
|
171 |
|
172 |
+
selected_prompt = st.selectbox('Select prompt', prompts)
|
|
|
173 |
|
174 |
+
items = items[items['prompt'] == selected_prompt].reset_index(drop=True)
|
175 |
+
prompt_id = items['prompt_id'].unique()[0]
|
176 |
+
note = items['note'].unique()[0]
|
|
|
177 |
|
178 |
+
# show source
|
179 |
+
if isinstance(note, str):
|
180 |
+
if note.isdigit():
|
181 |
+
st.caption(f"`Source: civitai`")
|
182 |
+
else:
|
183 |
+
st.caption(f"`Source: {note}`")
|
184 |
else:
|
185 |
+
st.caption("`Source: Parti-prompts`")
|
186 |
+
|
187 |
+
# show image metadata
|
188 |
+
image_metadatas = ['prompt_id', 'prompt', 'negativePrompt', 'sampler', 'cfgScale', 'size', 'seed']
|
189 |
+
for key in image_metadatas:
|
190 |
+
label = ' '.join(key.split('_')).capitalize()
|
191 |
+
st.write(f"**{label}**")
|
192 |
+
if items[key][0] == ' ':
|
193 |
+
st.write('`None`')
|
194 |
+
else:
|
195 |
+
st.caption(f"{items[key][0]}")
|
196 |
+
|
197 |
+
# for note as civitai image id, add civitai reference
|
198 |
+
if isinstance(note, str) and note.isdigit():
|
199 |
+
try:
|
200 |
+
st.write(f'**[Civitai Reference](https://civitai.com/images/{note})**')
|
201 |
+
res = requests.get(f'https://civitai.com/images/{note}')
|
202 |
+
# st.write(res.text)
|
203 |
+
soup = BeautifulSoup(res.text, 'html.parser')
|
204 |
+
image_section = soup.find('div', {'class': 'mantine-12rlksp'})
|
205 |
+
image_url = image_section.find('img')['src']
|
206 |
+
st.image(image_url, use_column_width=True)
|
207 |
+
except:
|
208 |
+
pass
|
209 |
+
|
210 |
+
return prompt_tags, tag, prompt_id, items
|
211 |
+
|
212 |
+
def app(self):
|
213 |
+
st.title('Model Visualization and Retrieval')
|
214 |
+
st.write('This is a gallery of images generated by the models')
|
215 |
+
|
216 |
+
prompt_tags, tag, prompt_id, items = self.sidebar()
|
217 |
+
|
218 |
+
# add safety check for some prompts
|
219 |
+
safety_check = True
|
220 |
+
unsafe_prompts = {}
|
221 |
+
# initialize unsafe prompts
|
222 |
+
for prompt_tag in prompt_tags:
|
223 |
+
unsafe_prompts[prompt_tag] = []
|
224 |
+
# manually add unsafe prompts
|
225 |
+
unsafe_prompts['world knowledge'] = [83]
|
226 |
+
# unsafe_prompts['art'] = [23]
|
227 |
+
unsafe_prompts['abstract'] = [1, 3]
|
228 |
+
# unsafe_prompts['food'] = [34]
|
229 |
+
|
230 |
+
if int(prompt_id.item()) in unsafe_prompts[tag]:
|
231 |
+
st.warning('This prompt may contain unsafe content. They might be offensive, depressing, or sexual.')
|
232 |
+
safety_check = st.checkbox('I understand that this prompt may contain unsafe content. Show these images anyway.', key=f'{prompt_id}')
|
233 |
+
|
234 |
+
if safety_check:
|
235 |
+
items, info, col_num = self.selection_panel(items)
|
236 |
+
|
237 |
+
if 'selected_dict' in st.session_state:
|
238 |
+
# st.write('checked: ', str(st.session_state.selected_dict.get(prompt_id, [])))
|
239 |
+
dynamic_weight_options = ['Grid Search', 'SVM', 'Greedy']
|
240 |
+
dynamic_weight_panel = st.columns(len(dynamic_weight_options))
|
241 |
+
|
242 |
+
if len(st.session_state.selected_dict.get(prompt_id, [])) > 0:
|
243 |
+
btn_disable = False
|
244 |
+
else:
|
245 |
+
btn_disable = True
|
246 |
+
|
247 |
+
for i in range(len(dynamic_weight_options)):
|
248 |
+
method = dynamic_weight_options[i]
|
249 |
+
with dynamic_weight_panel[i]:
|
250 |
+
btn = st.button(method, use_container_width=True, disabled=btn_disable, on_click=self.dynamic_weight, args=(prompt_id, items, method))
|
251 |
+
|
252 |
+
with st.form(key=f'{prompt_id}'):
|
253 |
+
# buttons = st.columns([1, 1, 1])
|
254 |
+
buttons_space = st.columns([1, 1, 1, 1])
|
255 |
+
gallery_space = st.empty()
|
256 |
+
|
257 |
+
with buttons_space[0]:
|
258 |
+
continue_btn = st.form_submit_button('Confirm Selection', use_container_width=True, type='primary')
|
259 |
+
if continue_btn:
|
260 |
+
self.submit_actions('Continue', prompt_id)
|
261 |
+
|
262 |
+
with buttons_space[1]:
|
263 |
+
select_btn = st.form_submit_button('Select All', use_container_width=True)
|
264 |
+
if select_btn:
|
265 |
+
self.submit_actions('Select', prompt_id)
|
266 |
+
|
267 |
+
with buttons_space[2]:
|
268 |
+
deselect_btn = st.form_submit_button('Deselect All', use_container_width=True)
|
269 |
+
if deselect_btn:
|
270 |
+
self.submit_actions('Deselect', prompt_id)
|
271 |
+
|
272 |
+
with buttons_space[3]:
|
273 |
+
refresh_btn = st.form_submit_button('Refresh', on_click=gallery_space.empty, use_container_width=True)
|
274 |
+
|
275 |
+
with gallery_space.container():
|
276 |
+
with st.spinner('Loading images...'):
|
277 |
+
self.gallery_standard(items, col_num, info)
|
278 |
+
|
279 |
+
prompt = st.chat_input(f"checked: {str(st.session_state.selected_dict.get(prompt_id, []))}", disabled=True, key=f'{prompt_id}')
|
280 |
+
|
281 |
+
def submit_actions(self, status, prompt_id):
|
282 |
+
if status == 'Select':
|
283 |
+
modelVersions = self.promptBook[self.promptBook['prompt_id'] == prompt_id]['modelVersion_id'].unique()
|
284 |
+
st.session_state.selected_dict[prompt_id] = modelVersions.tolist()
|
285 |
+
print(st.session_state.selected_dict, 'select')
|
286 |
+
st.experimental_rerun()
|
287 |
+
elif status == 'Deselect':
|
288 |
+
st.session_state.selected_dict[prompt_id] = []
|
289 |
+
print(st.session_state.selected_dict, 'deselect')
|
290 |
+
st.experimental_rerun()
|
291 |
+
# self.promptBook.loc[self.promptBook['prompt_id'] == prompt_id, 'checked'] = False
|
292 |
+
elif status == 'Continue':
|
293 |
+
st.session_state.selected_dict[prompt_id] = []
|
294 |
+
for key in st.session_state:
|
295 |
+
keys = key.split('_')
|
296 |
+
if keys[0] == 'select' and keys[1] == str(prompt_id):
|
297 |
+
if st.session_state[key]:
|
298 |
+
st.session_state.selected_dict[prompt_id].append(int(keys[2]))
|
299 |
+
# switch_page("ranking")
|
300 |
+
print(st.session_state.selected_dict, 'continue')
|
301 |
+
st.experimental_rerun()
|
302 |
+
|
303 |
+
def dynamic_weight(self, prompt_id, items, method='Grid Search'):
|
304 |
+
selected = items[
|
305 |
+
items['modelVersion_id'].isin(st.session_state.selected_dict[prompt_id])].reset_index(drop=True)
|
306 |
+
optimal_weight = [0, 0, 0]
|
307 |
+
|
308 |
+
if method == 'Grid Search':
|
309 |
+
# grid search method
|
310 |
+
top_ranking = len(items) * len(selected)
|
311 |
+
|
312 |
+
for clip_weight in np.arange(-1, 1, 0.1):
|
313 |
+
for mcos_weight in np.arange(-1, 1, 0.1):
|
314 |
+
for pop_weight in np.arange(-1, 1, 0.1):
|
315 |
+
|
316 |
+
weight_all = clip_weight*items[f'norm_clip'] + mcos_weight*items[f'norm_mcos'] + pop_weight*items['norm_pop']
|
317 |
+
weight_all_sorted = weight_all.sort_values(ascending=False).reset_index(drop=True)
|
318 |
+
# print('weight_all_sorted:', weight_all_sorted)
|
319 |
+
weight_selected = clip_weight*selected[f'norm_clip'] + mcos_weight*selected[f'norm_mcos'] + pop_weight*selected['norm_pop']
|
320 |
+
|
321 |
+
# get the index of values of weight_selected in weight_all_sorted
|
322 |
+
rankings = []
|
323 |
+
for weight in weight_selected:
|
324 |
+
rankings.append(weight_all_sorted.index[weight_all_sorted == weight].tolist()[0])
|
325 |
+
if sum(rankings) <= top_ranking:
|
326 |
+
top_ranking = sum(rankings)
|
327 |
+
print('current top ranking:', top_ranking, rankings)
|
328 |
+
optimal_weight = [clip_weight, mcos_weight, pop_weight]
|
329 |
+
print('optimal weight:', optimal_weight)
|
330 |
+
|
331 |
+
elif method == 'SVM':
|
332 |
+
# svm method
|
333 |
+
print('start svm method')
|
334 |
+
# get residual dataframe that contains models not selected
|
335 |
+
residual = items[~items['modelVersion_id'].isin(selected['modelVersion_id'])].reset_index(drop=True)
|
336 |
+
residual = residual[['norm_clip_crop', 'norm_mcos_crop', 'norm_pop']]
|
337 |
+
residual = residual.to_numpy()
|
338 |
+
selected = selected[['norm_clip_crop', 'norm_mcos_crop', 'norm_pop']]
|
339 |
+
selected = selected.to_numpy()
|
340 |
+
|
341 |
+
y = np.concatenate((np.full((len(selected), 1), -1), np.full((len(residual), 1), 1)), axis=0).ravel()
|
342 |
+
X = np.concatenate((selected, residual), axis=0)
|
343 |
+
|
344 |
+
# fit svm model, and get parameters for the hyperplane
|
345 |
+
clf = LinearSVC(random_state=0, C=1.0, fit_intercept=False, dual='auto')
|
346 |
+
clf.fit(X, y)
|
347 |
+
optimal_weight = clf.coef_[0].tolist()
|
348 |
+
print('optimal weight:', optimal_weight)
|
349 |
+
pass
|
350 |
+
|
351 |
+
elif method == 'Greedy':
|
352 |
+
for idx in selected.index:
|
353 |
+
# find which score is the highest, clip, mcos, or pop
|
354 |
+
clip_score = selected.loc[idx, 'norm_clip_crop']
|
355 |
+
mcos_score = selected.loc[idx, 'norm_mcos_crop']
|
356 |
+
pop_score = selected.loc[idx, 'norm_pop']
|
357 |
+
if clip_score >= mcos_score and clip_score >= pop_score:
|
358 |
+
optimal_weight[0] += 1
|
359 |
+
elif mcos_score >= clip_score and mcos_score >= pop_score:
|
360 |
+
optimal_weight[1] += 1
|
361 |
+
elif pop_score >= clip_score and pop_score >= mcos_score:
|
362 |
+
optimal_weight[2] += 1
|
363 |
+
|
364 |
+
# normalize optimal_weight
|
365 |
+
optimal_weight = [round(weight/len(selected), 2) for weight in optimal_weight]
|
366 |
+
print('optimal weight:', optimal_weight)
|
367 |
+
|
368 |
+
st.session_state.score_weights[0: 3] = optimal_weight
|
369 |
|
370 |
|
371 |
# hist_data = pd.DataFrame(np.random.normal(42, 10, (200, 1)), columns=["x"])
|
|
|
444 |
if 'selected_dict' not in st.session_state:
|
445 |
st.session_state['selected_dict'] = {}
|
446 |
|
447 |
+
app = GalleryApp(promptBook=promptBook, images_ds=images_ds)
|
448 |
+
app.app()
|
pages/__pycache__/Gallery.cpython-39.pyc
DELETED
Binary file (12.3 kB)
|
|
pages/streamlit-1.25.py
CHANGED
@@ -62,17 +62,17 @@ if st.button('Three cheers'):
|
|
62 |
if "chat_messages" not in st.session_state:
|
63 |
st.session_state.chat_messages = []
|
64 |
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
|
|
|
62 |
if "chat_messages" not in st.session_state:
|
63 |
st.session_state.chat_messages = []
|
64 |
|
65 |
+
prompt = st.chat_input("Say something")
|
66 |
+
if prompt:
|
67 |
+
st.session_state.chat_messages.append({"type": "user", "message": prompt})
|
68 |
+
st.session_state.chat_messages.append({"type": "bot", "message": "Hello!", "chart": np.random.randn(30, 3)})
|
69 |
+
|
70 |
+
for message in st.session_state.chat_messages[::-1]:
|
71 |
+
if message["type"] == "user":
|
72 |
+
with st.chat_message("You"):
|
73 |
+
st.write(message["message"])
|
74 |
+
else:
|
75 |
+
with st.chat_message("Bot"):
|
76 |
+
st.write(message["message"])
|
77 |
+
st.line_chart(message["chart"])
|
78 |
|