# tmp46541626387 — Data

来源：https://www.kaggle.com/competitions/gai-3006-01-00-final/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description



# CLEVR Train Set Summary

## 📦 1. Dataset

```

train/
├── images/                    (4,814개 이미지)
└── train_gt_labels.json       (학습용 정답 포함)


test/
└── images/                    (3,879개 이미지)
```

---

## 📘 2. Train GT lables JSON

* 경로: `train/train_gt_labels.json`

* `image_filename`: 연결된 이미지 파일
* `question`: 자연어 질문
* `answer`: 정답 (string)

```json
[
  {
    "image_filename": "train_770487.png",
    "question": "What color is the tiny matte block left of the blue block?",
    "answer": "gray"
  },
  {
    "image_filename": "train_670053.png",
    "question": "The large thing that is both on the left side of the purple shiny object and behind the tiny gray metallic ball is what color?",
    "answer": "brown"
  },
  {
    "image_filename": "train_216739.png",
    "question": "There is another tiny matte object that is the same shape as the tiny gray object; what color is it?",
    "answer": "cyan"
  },
....
]
```

---


## 📝 Submission Format

참가자가 제출해야 하는 형식:

```csv
id,answer
test_00000,brown
test_00001,yes
test_00002,purple
test_00003,3
...
test_06320,no
```

**요구사항:**
- CSV 파일
- 헤더: `id(파일이름),answer`
- 모든 3,879개 질문에 대한 답변 필요
- 답변 형식: 정확한 문자열 매칭 (대소문자 구분)
---