File size: 16,368 Bytes
43ef4e3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
311
312
313
314
315
316
317
318
319
320
321
322
323
[
  {
    "sample_id": "1",
    "category": "background_replacement",
    "total_tasks": 4,
    "tasks": [
      {
        "turn": 1,
        "modality": "text",
        "input": [
          {
            "text": "Observe the outfit of the person in the image, and choose the climate/weather environment from the options below that is most suitable for the person's current attire:\nA. A formal indoor gala or red carpet event\nB. A hot and humid tropical beach day\nC. A cold, snowy winter day\nD. A cool, breezy autumn evening\nE. None of the above",
            "image": "Apparel__background_replacement__1__img_0.jpg"
          }
        ],
        "output": {
          "answer": "A"
        }
      },
      {
        "turn": 2,
        "modality": "image",
        "input": [
          {
            "text": "Now, please modify the person's outfit based on the image from Turn 1. The modified outfit should be suitable for a cold, snowy winter day, while keeping all other unrelated features (including the person themselves, the image background, etc.) and the clothing's color, pattern, and style consistent. Only modify the clothing type reasonably as required."
          }
        ],
        "output": {
          "evaluation_points": [
            "The image should be clearly recognizable as a modification of the input image, not a completely newly generated image.",
            "The image's background (a seamless, uniform, bright white studio backdrop) must remain consistent with the Turn 1 input image, with no significant changes.",
            "The person's gender, facial appearance, hairstyle, body build, and dynamic pose must remain consistent with the Turn 1 input image, with no significant changes.",
            "The person's outfit must be suitable for a cold, snowy winter day (e.g., a long, insulated coat, warm pants, boots, gloves, scarf).",
            "The visual appearance of the person's entire outfit, including the color (iridescent teal/emerald green and black), pattern (intricate lace pattern), and high-fashion style, must remain consistent or highly similar to the outfit in the Turn 1 input image. The change should be in the clothing type/garment style, not its core visual aesthetic."
          ]
        }
      },
      {
        "turn": 3,
        "modality": "text",
        "input": [
          {
            "text": "Observe the background environment of the image you generated, and choose the scene type that best matches it:\nA. A professional photography studio with a seamless backdrop\nB. A busy city street at night\nC. A dense, sunlit forest\nD. A modern art gallery interior\nE. None of the above"
          }
        ],
        "output": {
          "answer": "A"
        }
      },
      {
        "turn": 4,
        "modality": "image",
        "input": [
          {
            "text": "Please move the person to a brand new scene based on your most recently generated image: a dense, sunlit forest. Only modify the background and keep all other unrelated elements unchanged."
          }
        ],
        "output": {
          "evaluation_points": [
            "The background must be completely replaced with a dense, sunlit forest scene, which should be clearly and distinctly different from the original uniform white studio backdrop. The new background's features (trees, foliage, sunlight filtering through) must be correct and prominent.",
            "The person's winter outfit (including its color, pattern, and style) must remain strictly consistent with the outfit in the Turn 2 output image. The outfit must also remain suitable for a cold, snowy winter day as required in Turn 2.",
            "The person's characteristics (face, hairstyle, body type) and dynamic pose must remain consistent with the Turn 1 input image, as these features have not been required to change in any turn.",
            "The person and their outfit must remain consistent with the Turn 2 output image, showing no significant changes other than the background replacement. The outfit must also remain suitable for a cold, snowy winter day as required in Turn 2.",
            "The image should be clearly recognizable as a modification of the Turn 2 output image, not a completely newly generated image."
          ]
        }
      }
    ],
    "system_prompt": "\nYou are an AI assistant participating in a multimodal benchmark evaluation. Strictly adhere to the following rules:\n\n1. **Multi-turn Multimodal Task**: This is a multi-turn task. You must synthesize all history information (including all prompt texts, input images, and your own previous text/image outputs) to determine your response strategy. Inputs may contain text and single or multiple images.\n\n2. **Output Modality Constraints**: Your response for each turn must be strictly EITHER \"Text\" OR \"Image\".\n   - If text output is required, treat it as a **Multiple Choice Question (MCQ)**.\n   - If image output is required, treat it as a **Generating Task**.\n\n3. **MCQ Guidelines**:\n   - All MCQs are **indeterminate choice** (can be single or multiple selection). Determine the number of correct options based on the prompt.\n   - **Output Format**: Output ONLY a contiguous string of uppercase letters representing the correct options, with no separators or punctuation (e.g., 'A' or 'BCD'). Do NOT output option content or explanations.\n\n4. **Generating Task Guidelines**:\n   - Generate one image based on all context.\n   - **Hint**: Unless explicitly instructed to \"generate a new image\" or \"modify a specific image,\" the images you generate should by default reference the most recent image in the context (whether input or previously generated by you) for further modification, or be re-created based on certain elements within it.",
    "domain": "Apparel",
    "subdomain": "background_replacement"
  },


  {
    "sample_id": "1",
    "category": "cooking_workflow",
    "total_tasks": 3,
    "tasks": [
      {
        "turn": 1,
        "modality": "text",
        "input": [
          {
            "text": "Observe the input dish image and select its correct name.\nOptions:\nA. Wonton\nB. Handmade Dumplings\nC. Spring Roll\nD. Steamed Bun\nE. None of the above.",
            "image": "Cooking__cooking_workflow__1__img_0.jpg"
          }
        ],
        "output": {
          "answer": "B"
        }
      },
      {
        "turn": 2,
        "modality": "text",
        "input": [
          {
            "text": "In the process of making handmade dumplings, which of the following steps is an essential core procedure?\nOptions:\nA. Rolling out the dumpling wrappers\nB. Deep-frying the filling\nC. Baking the dough\nD. Grilling the meat\nE. None of the above."
          }
        ],
        "output": {
          "answer": "A"
        }
      },
      {
        "turn": 3,
        "modality": "image",
        "input": [
          {
            "text": "Following the correct step selected in the previous round, the next step that typically follows immediately is one of the options below. Please select the correct option and then directly generate the operational scene of this step (do not output the option text)."
          }
        ],
        "output": {
          "evaluation_points": [
            "The image depicts the specific operational scene of 'wrapping the filling and pinching the pleats'.",
            "The ingredients shown in the scene (such as the dumpling wrappers and filling) match their state at the 'wrapping the filling and pinching the pleats' stage of the process (including form, doneness, color, etc.), consistent with their depiction in the input image from turn 1.",
            "The cooking action is correct and logically consistent."
          ]
        }
      }
    ],
    "system_prompt": "<SYSTEM_PROMPT_PLACEHOLDER>",
    "domain": "Cooking",
    "subdomain": "cooking_workflow"
  },


  {
    "sample_id": "1",
    "category": "room_renovation",
    "total_tasks": 2,
    "tasks": [
      {
        "turn": 1,
        "modality": "text",
        "input": [
          {
            "text": "This is an image of an unfurnished room. If this room were to be renovated in a modern minimalist style and used as a bathroom, which of the following items should be present in the room?\nOptions:\nA. A four-poster canopy bed\nB. A wall-mounted toilet\nC. A large wooden dining table\nD. A floor-to-ceiling bookshelf\nE. None of the above.",
            "image": "Interior_Design__room_renovation__1__img_0.jpg"
          }
        ],
        "output": {
          "answer": "B"
        }
      },
      {
        "turn": 2,
        "modality": "image",
        "input": [
          {
            "text": "Please renovate the image of this unfurnished room in the modern minimalist style mentioned in the previous turn. Complete the hard finishes for the walls and floor, and add the furniture item you selected in the previous turn's options."
          }
        ],
        "output": {
          "evaluation_points": [
            "The unfinished walls and floor are covered with finishes (e.g., paint, tiles, flooring) appropriate for a modern minimalist style.",
            "A wall-mounted toilet is added to the room, with its features clearly visible and placed in a logical position (e.g., near the plumbing connections and floor drain).",
            "The room's hard structural elements (spatial layout, positions of the water pipe inlets, floor drain, and the absence of windows) remain strictly consistent with the Turn 1 input image."
          ]
        }
      }
    ],
    "system_prompt": "<SYSTEM_PROMPT_PLACEHOLDER>",
    "domain": "Interior_Design",
    "subdomain": "room_renovation"
  },


  {
    "sample_id": "1",
    "category": "elastic_storage",
    "total_tasks": 4,
    "tasks": [
      {
        "turn": 1,
        "modality": "image",
        "input": [
          {
            "text": "Generate a basketball in its fully inflated, round state. The background should be plain white."
          }
        ],
        "output": {
          "evaluation_points": [
            "The image correctly depicts a basketball with its typical visual characteristics (e.g., orange color, black lines, pebbled texture).",
            "The basketball is shown in its fully inflated, spherical state, maximizing its volume.",
            "The image background is a simple, plain white."
          ]
        }
      },
      {
        "turn": 2,
        "modality": "image",
        "input": [
          {
            "text": "Based on the basketball image from Turn 1, please show it being stored inside a standard student laptop backpack. The backpack should be open, allowing both the backpack and the stored item to be visible. Consider the backpack's size and space constraints. Do not damage or destroy any item."
          }
        ],
        "output": {
          "evaluation_points": [
            "The basketball is successfully placed inside the backpack. The backpack is open, and both the backpack and the basketball are visible.",
            "The basketball is shown as being stored via deflation and flattening, not by magically shrinking it to its original shape.",
            "The backpack itself is not deformed, stretched, or torn.",
            "The background and other elements are consistent with the reference image from Turn 1.",
            "The basketball, although deflated and flattened, retains its typical characteristics (e.g., orange color, black lines, texture)."
          ]
        }
      },
      {
        "turn": 3,
        "modality": "text",
        "input": [
          {
            "text": "Based on the images you generated, what operation was performed on the basketball to successfully store it in the backpack?\n\nOptions:\nA. Cutting it into pieces\nB. Completely deflating and flattening it\nC. Shrinking it magically to a smaller size\nD. Melting it into a liquid\nE. None of the above."
          }
        ],
        "output": {
          "answer": "B"
        }
      },
      {
        "turn": 4,
        "modality": "text",
        "input": [
          {
            "text": "Please check the image generated in Turn 2. Is the basketball completely contained within the backpack, and is the backpack itself undamaged?\n\nOptions:\nA. Yes, both conditions are fully met.\nB. No, the task was not completed properly.\nE. None of the above."
          }
        ],
        "output": {
          "answer": "<DYNAMIC>"
        }
      }
    ],
    "system_prompt": "<SYSTEM_PROMPT_PLACEHOLDER>",
    "domain": "Storage",
    "subdomain": "elastic_storage"
  },


  {
    "sample_id": "1",
    "category": "stack_word_formation",
    "total_tasks": 5,
    "tasks": [
      {
        "turn": 1,
        "modality": "image",
        "input": [
          {
            "text": "Generate a single horizontal row of 5 equal-sized, contiguous squares (white background, black border) to represent a stack. The leftmost square is the bottom of the stack, and the right is the top direction. The stack is currently empty."
          }
        ],
        "output": {
          "evaluation_points": [
            "The image contains exactly one row of 5 equal-sized squares, arranged horizontally and contiguously.",
            "All squares are completely empty inside, containing no characters, patterns, or fill.",
            "The background is clean and simple, highlighting the grid structure."
          ]
        }
      },
      {
        "turn": 2,
        "modality": "text",
        "input": [
          {
            "text": "Into the empty stack shown in Turn 1, push the characters R, I, C in order from bottom to top, then immediately pop one character. Which character is popped?\nOptions:\nA. R\nB. I\nC. C\nD. H\nE. None of the above."
          }
        ],
        "output": {
          "answer": "C"
        }
      },
      {
        "turn": 3,
        "modality": "image",
        "input": [
          {
            "text": "Based on the stack state after the pop operation in Turn 2, now push the characters A and H in that order. Using the empty stack you drew in Turn 1, generate the current state of the stack."
          }
        ],
        "output": {
          "evaluation_points": [
            "Maintain the same 5 horizontal, contiguous squares (white background, black border) as specified in Turn 1's instruction ('a single horizontal row of 5 equal-sized, contiguous squares (white background, black border)'), with the stack bottom on the left.",
            "The stack should contain 4 characters: R, I, A, H, occupying the first 4 squares from left to right, in that order.",
            "The rightmost (5th) square must remain empty.",
            "Each character is centered within its own square, is clear and readable, and does not overlap or extend outside the border."
          ]
        }
      },
      {
        "turn": 4,
        "modality": "text",
        "input": [
          {
            "text": "Starting from the stack state shown in Turn 3, pop all characters from the stack until it is empty. In this popping process, what is the 3rd character popped?\nOptions:\nA. R\nB. I\nC. C\nD. A\nE. H\nF. None of the above."
          }
        ],
        "output": {
          "answer": "B"
        }
      },
      {
        "turn": 5,
        "modality": "image",
        "input": [
          {
            "text": "Consider all the letters popped in this problem. If you combine them in the order they were popped, you get a 5-letter English word. On a pure white background, generate the object represented by this English word (do not write the word itself)."
          }
        ],
        "output": {
          "evaluation_points": [
            "The image depicts the object represented by the 5-letter English word formed by combining the popped letters 'C', 'H', 'A', 'I', 'R' in order, which is a chair. Its appearance is highly consistent with the typical representation of a chair.",
            "The image background is pure white or nearly pure white, with no extra scenery or props causing distraction.",
            "No text appears anywhere in the image."
          ]
        }
      }
    ],
    "system_prompt": "<SYSTEM_PROMPT_PLACEHOLDER>",
    "domain": "Stack_Simulation",
    "subdomain": "stack_word_formation"
  }
]