Emilyxml commited on
Commit
29de8cf
·
verified ·
1 Parent(s): cff5e86

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +158 -201
app.py CHANGED
@@ -8,13 +8,13 @@ from pathlib import Path
8
  from huggingface_hub import CommitScheduler
9
 
10
  # --- 1. 配置区域 ---
11
- DATASET_REPO_ID = "Emilyxml/moveit" # 你的数据集
12
- DATA_FOLDER = "data" # 数据文件夹
13
- LOG_FOLDER = Path("logs") # 本地日志
14
  LOG_FOLDER.mkdir(parents=True, exist_ok=True)
15
  TOKEN = os.environ.get("HF_TOKEN")
16
 
17
- # --- 2. 启动同步调度器 ---
18
  scheduler = CommitScheduler(
19
  repo_id=DATASET_REPO_ID,
20
  repo_type="dataset",
@@ -24,7 +24,7 @@ scheduler = CommitScheduler(
24
  token=TOKEN
25
  )
26
 
27
- # --- 3. 数据加载逻辑 (区分原图和候选图) ---
28
  def load_data():
29
  groups = {}
30
  if not os.path.exists(DATA_FOLDER):
@@ -34,21 +34,16 @@ def load_data():
34
  for filename in os.listdir(DATA_FOLDER):
35
  if filename.startswith('.'): continue
36
  file_path = os.path.join(DATA_FOLDER, filename)
37
- prefix = filename[:5] # 以前5个字符作为组ID
38
 
39
  if prefix not in groups:
40
- # origin 存原图路径,candidates 存其他方法的图
41
  groups[prefix] = {"origin": None, "candidates": [], "instruction": "暂无说明"}
42
 
43
- # 图片处理
44
  if filename.lower().endswith(('.png', '.jpg', '.jpeg', '.webp')):
45
- # 判断是否是原图 (文件名包含 _origin)
46
  if "_origin" in filename.lower():
47
  groups[prefix]["origin"] = file_path
48
  else:
49
  groups[prefix]["candidates"].append(file_path)
50
-
51
- # 文本处理
52
  elif filename.lower().endswith('.txt'):
53
  try:
54
  with open(file_path, "r", encoding="utf-8") as f:
@@ -57,7 +52,6 @@ def load_data():
57
  with open(file_path, "r", encoding="gbk") as f:
58
  groups[prefix]["instruction"] = f.read()
59
 
60
- # 过滤掉既没有原图也没有候选图的组
61
  valid_groups = {}
62
  for k, v in groups.items():
63
  if v["origin"] is not None or len(v["candidates"]) > 0:
@@ -70,18 +64,11 @@ def load_data():
70
 
71
  ALL_GROUPS, ALL_GROUP_IDS = load_data()
72
 
73
- # --- 4. 保存逻辑 ---
74
  def save_user_vote(user_id, group_id, choice_labels, method_names):
75
  user_filename = f"user_{user_id}.csv"
76
  user_file_path = LOG_FOLDER / user_filename
77
-
78
- row = [
79
- user_id,
80
- datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
81
- group_id,
82
- choice_labels,
83
- method_names
84
- ]
85
 
86
  with scheduler.lock:
87
  file_exists = user_file_path.exists()
@@ -90,216 +77,186 @@ def save_user_vote(user_id, group_id, choice_labels, method_names):
90
  if not file_exists:
91
  writer.writerow(["user_id", "timestamp", "group_id", "selected_labels", "selected_methods"])
92
  writer.writerow(row)
93
-
94
- print(f"Saved: User {user_id} selected {method_names}")
95
 
96
- # --- 5. 交互逻辑 ---
97
 
98
- def get_current_question_ui(user_state):
99
- """刷新界面:原图置顶 + 候选图乱序"""
100
- current_idx = user_state["index"]
101
-
102
- if current_idx >= len(ALL_GROUP_IDS):
103
- return (
104
- gr.update(visible=False), gr.update(visible=False), gr.update(visible=False),
105
- gr.update(visible=False), gr.update(value="## 🎉 测试结束!\n感谢您的参与。", visible=True),
106
- user_state, [], [], False # False表示当前没有原图显示
107
- )
 
 
 
 
108
 
109
- group_id = ALL_GROUP_IDS[current_idx]
 
 
 
 
110
  group_data = ALL_GROUPS[group_id]
111
 
112
- instruction_text = f"### 任务 ({current_idx + 1} / {len(ALL_GROUP_IDS)})\n\n{group_data['instruction']}"
 
113
 
114
- # --- 构建显示列表 ---
115
- display_list = []
116
- real_paths = [] # 记录真实的路径顺序
117
-
118
- # 1. 先放原图 (如果有)
119
- has_origin = False
120
- if group_data["origin"]:
121
- display_list.append((group_data["origin"], "Reference (原图)"))
122
- real_paths.append(group_data["origin"])
123
- has_origin = True
124
-
125
- # 2. 再放候选图 (打乱顺序)
126
  candidates = group_data["candidates"].copy()
127
  random.shuffle(candidates)
128
 
129
- for i, img_path in enumerate(candidates):
130
- # 标签从 Option A 开始
131
- label = f"Option {chr(65+i)}"
132
- display_list.append((img_path, label))
133
- real_paths.append(img_path)
134
-
135
- # 动态计算列数 (原图+候选图的总数)
136
- total_imgs = len(display_list)
137
- # 如果总数是2(1原图+1候选),显示2列
138
- # 如果总数是3,显示3列
139
- # 如果总数是4或以上,显示3列换行,或者你可以改成 4
140
- cols = min(total_imgs, 3)
141
 
142
- return (
143
- gr.update(value=instruction_text, visible=True),
144
- gr.update(value=display_list, columns=cols, visible=True),
145
- gr.update(value="当前未选择任何图片", visible=True),
146
- gr.update(visible=True),
147
- gr.update(visible=False),
148
- user_state,
149
- real_paths, # 存下当前界面所有图片的真实路径
150
- [], # 清空选中项
151
- has_origin # 告诉前端第一张是不是原图
152
- )
153
 
154
- def toggle_selection(evt: gr.SelectData, current_indices, has_origin):
155
- """
156
- 处理点击。
157
- has_origin: 如果为 True,说明 Index 0 是原图,禁止选中。
158
- """
159
- clicked_idx = evt.index
160
-
161
- # --- 保护机制:如果第一张是原图,点击无效 ---
162
- if has_origin and clicked_idx == 0:
163
- # 你可以去掉这个if,允许选原图,但通常 User Study 不需要选 Reference
164
- # 这里我们返回原样,不更新选中状态
165
- return current_indices, "⚠️ 您点击的是参考原图 (Reference),请选择后面的选项。"
166
-
167
- # 切换选中状态
168
- if clicked_idx in current_indices:
169
- current_indices.remove(clicked_idx)
170
- else:
171
- current_indices.append(clicked_idx)
172
- current_indices.sort()
173
-
174
- # 生成状态文本
175
- if not current_indices:
176
- status_text = "当前未选择任何图片"
177
- else:
178
- # 计算显示的 Label
179
- labels = []
180
- for idx in current_indices:
181
- # 如果有原图,index 1 才是 Option A
182
- if has_origin:
183
- option_char = chr(65 + (idx - 1)) # idx=1 -> A, idx=2 -> B
184
- else:
185
- option_char = chr(65 + idx)
186
- labels.append(f"Option {option_char}")
187
-
188
- status_text = "已选中: " + ", ".join(labels)
189
-
190
- return current_indices, status_text
191
 
192
- def submit_vote(user_state, current_file_paths, current_indices, is_none=False, has_origin=False):
193
- user_id = user_state["user_id"]
194
  current_idx = user_state["index"]
195
-
196
- if current_idx >= len(ALL_GROUP_IDS):
197
- return get_current_question_ui(user_state)
198
-
199
  group_id = ALL_GROUP_IDS[current_idx]
200
-
 
201
  if is_none:
202
- save_user_vote(user_id, group_id, "Rejected All", "None_Satisfied")
203
- user_state["index"] += 1
204
- return get_current_question_ui(user_state)
205
-
206
- if not current_indices:
207
- return (
208
- gr.update(), gr.update(),
209
- gr.update(value="❌ 请至少选择一张图片"),
210
- gr.update(), gr.update(),
211
- user_state, current_file_paths, current_indices, has_origin
212
- )
213
-
214
- selected_labels = []
215
- selected_methods = []
216
-
217
- for idx in current_indices:
218
- # 获取真实路径
219
- real_path = current_file_paths[idx]
220
-
221
- # 计算 Label (Option A/B...)
222
- if has_origin:
223
- label = f"Option {chr(65 + (idx - 1))}"
224
- else:
225
- label = f"Option {chr(65 + idx)}"
226
- selected_labels.append(label)
227
-
228
- # 提取方法名
229
- filename = os.path.basename(real_path)
230
- name_no_ext = os.path.splitext(filename)[0]
231
-
232
- # 如果是原图被选中了(假设你去掉了保护机制),方法名就是 origin
233
- if "_origin" in name_no_ext:
234
- method = "reference_origin"
235
- else:
236
  parts = name_no_ext.split('_', 1)
237
  method = parts[1] if len(parts) > 1 else name_no_ext
 
238
 
239
- selected_methods.append(method)
240
-
241
- str_labels = "; ".join(selected_labels)
242
- str_methods = "; ".join(selected_methods)
243
-
244
- save_user_vote(user_id, group_id, str_labels, str_methods)
245
-
246
  user_state["index"] += 1
247
- return get_current_question_ui(user_state)
 
 
248
 
249
- # --- 6. 界面构建 ---
250
  with gr.Blocks(title="User Study") as demo:
251
 
252
- state_user = gr.State(lambda: {"user_id": str(uuid.uuid4())[:8], "index": 0})
253
- state_files = gr.State([])
254
- state_indices = gr.State([])
255
- state_has_origin = gr.State(False) # 记录当前页面是否有原图
 
 
 
 
 
 
 
 
256
 
257
- with gr.Column():
258
- instruction_md = gr.Markdown("Loading...")
 
259
 
260
- # 图片显示
261
- gallery = gr.Gallery(
262
- label="请参考第一张原图,选择后面最好的结果(可多选)",
263
- allow_preview=True,
264
- object_fit="contain",
265
- height="auto",
266
- interactive=True
267
- )
268
-
269
- status_box = gr.Textbox(value="当前未选择任何图片", label="当前选中状态", interactive=False)
270
-
271
- with gr.Row():
272
- btn_submit = gr.Button("✅ 提交选择", variant="primary", scale=2)
273
- btn_none = gr.Button("🚫 都不满意", variant="stop", scale=1)
274
 
275
- end_msg = gr.Markdown(visible=False)
 
 
 
 
 
 
276
 
277
- # 加载第一题
278
- demo.load(
279
- fn=get_current_question_ui,
280
- inputs=[state_user],
281
- outputs=[instruction_md, gallery, status_box, btn_submit, end_msg, state_user, state_files, state_indices, state_has_origin]
282
- )
283
-
284
- # 点击图片
285
- gallery.select(
286
- fn=toggle_selection,
287
- inputs=[state_indices, state_has_origin],
288
- outputs=[state_indices, status_box]
289
- )
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
290
 
291
- # 提交按钮
 
 
 
 
 
 
 
 
 
 
 
 
 
292
  btn_submit.click(
293
- fn=lambda s, f, i, h: submit_vote(s, f, i, is_none=False, has_origin=h),
294
- inputs=[state_user, state_files, state_indices, state_has_origin],
295
- outputs=[instruction_md, gallery, status_box, btn_submit, end_msg, state_user, state_files, state_indices, state_has_origin]
296
  )
297
-
298
- # 都不满意按钮
 
 
 
 
 
299
  btn_none.click(
300
- fn=lambda s, f, i, h: submit_vote(s, f, i, is_none=True, has_origin=h),
301
- inputs=[state_user, state_files, state_indices, state_has_origin],
302
- outputs=[instruction_md, gallery, status_box, btn_submit, end_msg, state_user, state_files, state_indices, state_has_origin]
303
  )
304
 
305
  if __name__ == "__main__":
 
8
  from huggingface_hub import CommitScheduler
9
 
10
  # --- 1. 配置区域 ---
11
+ DATASET_REPO_ID = "Emilyxml/moveit"
12
+ DATA_FOLDER = "data"
13
+ LOG_FOLDER = Path("logs")
14
  LOG_FOLDER.mkdir(parents=True, exist_ok=True)
15
  TOKEN = os.environ.get("HF_TOKEN")
16
 
17
+ # --- 2. 启动同步 ---
18
  scheduler = CommitScheduler(
19
  repo_id=DATASET_REPO_ID,
20
  repo_type="dataset",
 
24
  token=TOKEN
25
  )
26
 
27
+ # --- 3. 数据加载 (不变) ---
28
  def load_data():
29
  groups = {}
30
  if not os.path.exists(DATA_FOLDER):
 
34
  for filename in os.listdir(DATA_FOLDER):
35
  if filename.startswith('.'): continue
36
  file_path = os.path.join(DATA_FOLDER, filename)
37
+ prefix = filename[:5]
38
 
39
  if prefix not in groups:
 
40
  groups[prefix] = {"origin": None, "candidates": [], "instruction": "暂无说明"}
41
 
 
42
  if filename.lower().endswith(('.png', '.jpg', '.jpeg', '.webp')):
 
43
  if "_origin" in filename.lower():
44
  groups[prefix]["origin"] = file_path
45
  else:
46
  groups[prefix]["candidates"].append(file_path)
 
 
47
  elif filename.lower().endswith('.txt'):
48
  try:
49
  with open(file_path, "r", encoding="utf-8") as f:
 
52
  with open(file_path, "r", encoding="gbk") as f:
53
  groups[prefix]["instruction"] = f.read()
54
 
 
55
  valid_groups = {}
56
  for k, v in groups.items():
57
  if v["origin"] is not None or len(v["candidates"]) > 0:
 
64
 
65
  ALL_GROUPS, ALL_GROUP_IDS = load_data()
66
 
67
+ # --- 4. 保存逻辑 (不变) ---
68
  def save_user_vote(user_id, group_id, choice_labels, method_names):
69
  user_filename = f"user_{user_id}.csv"
70
  user_file_path = LOG_FOLDER / user_filename
71
+ row = [user_id, datetime.now().strftime("%Y-%m-%d %H:%M:%S"), group_id, choice_labels, method_names]
 
 
 
 
 
 
 
72
 
73
  with scheduler.lock:
74
  file_exists = user_file_path.exists()
 
77
  if not file_exists:
78
  writer.writerow(["user_id", "timestamp", "group_id", "selected_labels", "selected_methods"])
79
  writer.writerow(row)
80
+ print(f"Saved: {user_id} -> {choice_labels}")
 
81
 
82
+ # --- 5. 核心逻辑:状态管理 ---
83
 
84
+ def init_state():
85
+ """初始化用户状态"""
86
+ return {
87
+ "user_id": str(uuid.uuid4())[:8],
88
+ "index": 0,
89
+ "is_finished": False
90
+ }
91
+
92
+ def next_question_data(user_state):
93
+ """
94
+ 计算下一题的数据,并返回给前端渲染。
95
+ 这里只处理数据准备(打乱顺序等),不处理 UI。
96
+ """
97
+ idx = user_state["index"]
98
 
99
+ if idx >= len(ALL_GROUP_IDS):
100
+ user_state["is_finished"] = True
101
+ return user_state, None, [], [] # 结束状态
102
+
103
+ group_id = ALL_GROUP_IDS[idx]
104
  group_data = ALL_GROUPS[group_id]
105
 
106
+ # 准备原图
107
+ origin_path = group_data["origin"]
108
 
109
+ # 准备候选图(打乱)
 
 
 
 
 
 
 
 
 
 
 
110
  candidates = group_data["candidates"].copy()
111
  random.shuffle(candidates)
112
 
113
+ # 构造候选图信息列表 [(path, "Option A"), (path, "Option B")...]
114
+ candidate_info = []
115
+ for i, path in enumerate(candidates):
116
+ label = f"Option {chr(65+i)}"
117
+ candidate_info.append({"path": path, "label": label})
 
 
 
 
 
 
 
118
 
119
+ return user_state, origin_path, candidate_info, group_data["instruction"]
 
 
 
 
 
 
 
 
 
 
120
 
121
+ def submit_logic(user_state, current_candidates, selected_indices, is_none=False):
122
+ """处理提交,保存数据,并进入下一题"""
123
+ if user_state["is_finished"]:
124
+ return user_state, [], []
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
125
 
 
 
126
  current_idx = user_state["index"]
 
 
 
 
127
  group_id = ALL_GROUP_IDS[current_idx]
128
+
129
+ # 1. 保存数据
130
  if is_none:
131
+ save_user_vote(user_state["user_id"], group_id, "Rejected All", "None_Satisfied")
132
+ else:
133
+ # 如果不是 None 但没选,直接返回不跳转(或者你可以允许)
134
+ if not selected_indices:
135
+ raise gr.Error("请至少选择一张图片,或点击“都不满意”")
136
+
137
+ labels = []
138
+ methods = []
139
+ for idx in selected_indices:
140
+ info = current_candidates[idx]
141
+ labels.append(info["label"])
142
+
143
+ # 提取方法名
144
+ filename = os.path.basename(info["path"])
145
+ name_no_ext = os.path.splitext(filename)[0]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
146
  parts = name_no_ext.split('_', 1)
147
  method = parts[1] if len(parts) > 1 else name_no_ext
148
+ methods.append(method)
149
 
150
+ save_user_vote(user_state["user_id"], group_id, "; ".join(labels), "; ".join(methods))
151
+
152
+ # 2. 索引+1
 
 
 
 
153
  user_state["index"] += 1
154
+
155
+ # 3. 清空选择
156
+ return user_state, [] # 返回新的 state 和空的 selected_indices
157
 
158
+ # --- 6. 界面构建 (使用 @gr.render 实现自定义布局) ---
159
  with gr.Blocks(title="User Study") as demo:
160
 
161
+ # === 状态变量 ===
162
+ # state_main: 存 user_id, index
163
+ state_main = gr.State(init_state())
164
+ # state_current_data: 存当前题目的数据 (原图, 候选图列表, 说明文)
165
+ state_origin = gr.State()
166
+ state_candidates = gr.State([]) # list of dict
167
+ state_instruction = gr.State("")
168
+ # state_selection: 存当前选中的索引 [0, 2]
169
+ state_selection = gr.State([])
170
+
171
+ # === 页面布局 ===
172
+ header = gr.Markdown("Loading...")
173
 
174
+ # 动态渲染区域
175
+ @gr.render(inputs=[state_main, state_origin, state_candidates, state_selection, state_instruction])
176
+ def render_content(main_st, origin, candidates, selection, instruction):
177
 
178
+ # 1. 如果结束了
179
+ if main_st["is_finished"]:
180
+ gr.Markdown("## 🎉 测试结束!\n感谢您的参与,所有结果已保存。")
181
+ return
182
+
183
+ # 2. 显示说明
184
+ idx = main_st["index"]
185
+ gr.Markdown(f"### 任务 ({idx + 1} / {len(ALL_GROUP_IDS)})\n\n{instruction}")
 
 
 
 
 
 
186
 
187
+ # 3. 显示原图 (Reference)
188
+ if origin:
189
+ with gr.Row():
190
+ with gr.Column(scale=1):
191
+ gr.Image(origin, label="Reference (参考原图)", interactive=False, height=300)
192
+ with gr.Column(scale=2):
193
+ gr.Markdown("👈 **请参考左侧原图**,并在下方选择您认为质量最好的图片(可多选)。\n\n**点击图片下方的按钮进行选择。**")
194
 
195
+ # 4. 显示候选图 (Grid Layout)
196
+ # 使用 Row wrap=True 实现自动换行
197
+ with gr.Row(wrap=True):
198
+ for i, item in enumerate(candidates):
199
+ is_selected = i in selection
200
+
201
+ # 定义每个卡片的样式
202
+ with gr.Column(min_width=200): # 限制最小宽度,类似 Gallery 效果
203
+ gr.Image(item["path"], show_label=False, interactive=False)
204
+
205
+ # === 核心:每个图片下的按钮 ===
206
+ btn_text = f"✅ {item['label']} (已选)" if is_selected else f"⬜️ {item['label']} (点击选择)"
207
+ btn_variant = "primary" if is_selected else "secondary"
208
+
209
+ btn = gr.Button(btn_text, variant=btn_variant)
210
+
211
+ # 按钮点击逻辑:切换选中状态
212
+ def toggle(idx, current_sel):
213
+ if idx in current_sel:
214
+ current_sel.remove(idx)
215
+ else:
216
+ current_sel.append(idx)
217
+ current_sel.sort()
218
+ return current_sel
219
+
220
+ # 绑定点击事件,更新 state_selection,从而触发重绘
221
+ btn.click(fn=toggle, inputs=[gr.Number(i, visible=False), state_selection], outputs=[state_selection])
222
+
223
+ # === 底部操作栏 ===
224
+ with gr.Row():
225
+ btn_submit = gr.Button("🚀 提交 (Submit)", variant="primary", scale=2)
226
+ btn_none = gr.Button("🚫 都不满意 (None)", variant="stop", scale=1)
227
+
228
+ # === 事件流 ===
229
 
230
+ # 1. 初始化加载第一题
231
+ def load_first(main_st):
232
+ return next_question_data(main_st)
233
+
234
+ demo.load(load_first, inputs=[state_main], outputs=[state_main, state_origin, state_candidates, state_instruction])
235
+
236
+ # 2. 提交按钮 -> 保存 -> 准备下一题 -> 清空选择
237
+ def on_submit(main_st, cands, sel):
238
+ # 先保存
239
+ new_main, new_sel = submit_logic(main_st, cands, sel, is_none=False)
240
+ # 再加载下一题数据
241
+ updated_main, origin, new_cands, instr = next_question_data(new_main)
242
+ return updated_main, new_sel, origin, new_cands, instr
243
+
244
  btn_submit.click(
245
+ fn=on_submit,
246
+ inputs=[state_main, state_candidates, state_selection],
247
+ outputs=[state_main, state_selection, state_origin, state_candidates, state_instruction]
248
  )
249
+
250
+ # 3. 都不满意 -> 保存 -> 准备下一题
251
+ def on_none(main_st, cands, sel):
252
+ new_main, new_sel = submit_logic(main_st, cands, sel, is_none=True)
253
+ updated_main, origin, new_cands, instr = next_question_data(new_main)
254
+ return updated_main, new_sel, origin, new_cands, instr
255
+
256
  btn_none.click(
257
+ fn=on_none,
258
+ inputs=[state_main, state_candidates, state_selection],
259
+ outputs=[state_main, state_selection, state_origin, state_candidates, state_instruction]
260
  )
261
 
262
  if __name__ == "__main__":