| --- |
| license: other |
| task_categories: |
| - visual-question-answering |
| language: |
| - en |
| --- |
| |
| # est_vqa_cleaned |
|
|
| The **est_vqa__x** family of the ElliotVL supervised-fine-tuning pool, **after VLM cleaning**. |
| |
| | | | |
| |---|---| |
| | images | 16,867 | |
| | QA turns | 77,754 | |
| | answers rewritten by the cleaning pass | 0 | |
| | QA created by the cleaning pass (`new_qa`) | not measured for this family | |
| | shards | 32 | |
| |
| ## How this was cleaned |
| |
| A vision-language model read each image together with its QA and judged the item. The pass is |
| **not a filter that only removes rows** — it rewrites answers it finds wrong but salvageable, |
| drops what it cannot salvage, and adds QA where the image supports more than the source provided. |
| Each row carries the judge's own record in `clean_meta`, including the cleaner identity, the |
| policy it applied, and its per-item scores for legibility, richness and coverage. |
|
|
| **A large share of the QA here was written by the cleaning pass, not by the original dataset.** |
| Across the pool that share runs from roughly half to over 80% of a family's turns, and it is |
| reported in the table above. Those rows have an empty `org_answer` because no pre-clean original |
| exists. Treat them as model-generated supervision: they were judged against the image, but they |
| are not human annotation, and model-written QA is where formatting defects are most likely. |
|
|
| The effect on the answers that *were* carried over is substantive rather than cosmetic. In this pool the pass corrects values, not just |
| wording — a curve's critical point restated from `4.00` to `2.00`, a computed ratio from `1` to |
| `Approximately 1.33` — and for some families it removed the content entirely, which is why those |
| families are absent here rather than published empty. |
|
|
| ## `answer` vs `org_answer` |
| |
| - `answer` — the cleaned answer. **Train on this.** |
| - `org_answer` — the pre-cleaning answer from the same `(image, question)` in the uncleaned pool. |
| It is empty for QA the cleaning pass added, which have no original. |
|
|
| > For this family `org_answer` is empty throughout: recovering it means scanning the whole pre-clean family, which is raw family is 16 GB (> 6 GB cap). The cleaned `answer` is unaffected. |
| |
| Where the two differ, the difference is the correction. Keeping both makes every edit auditable |
| instead of asking you to take the cleaning on trust. |
| |
| |
| |
| ## Schema |
| |
| - `image` — HF `Image()`, renders directly in the dataset viewer |
| - `image_sha256` — content hash of the image |
| - `n_qa` — number of turns attached to this image |
| - `qa` — list of {`question`, `answer`, `org_answer`}, **in source order** |
| - `family` / `clean_meta` — provenance and the judge's per-row record |
|
|
| A row with more than one entry in `qa` is a **conversation over a single image**, not a set of |
| independent questions — the turns depend on each other and the image belongs to all of them. |
| Keep them together and in order when training. |
|
|
| Parquet row groups are written at ~32 MB so the dataset viewer pages without stalling. |
|
|
| ## What is NOT claimed |
|
|
| No rejection sampling was run against this family: there is no accept/reject verdict per row, and |
| `answer` is the cleaned reference rather than a model generation that a string verifier confirmed. |
| For that, use the `*_RS_think` / `*_rejected` datasets in [elliot-mllm](https://huggingface.co/elliot-mllm). |
|
|
|
|
| ## Examples |
| 10 rows taken straight from this dataset. Images are the **original files** as stored here - no downscaling, no recompression - and every turn is shown in full. |
|
|
| ### Example 1 — 5 turns on one image |
|
|
|  |
|
|
| | # | question | answer | |
| |---|---|---| |
| | 1 | 这家店的名字是什么? | 禾不锈钢 | |
| | 2 | 招牌右侧除了店名还写了什么具体的服务项目? | 6米剪板折弯 | |
| | 3 | 招牌下方列出了哪三个联系电话号码? | 18502304582, 18996293017, 18523195639 | |
| | 4 | 店铺门口上方安装了什么颜色的卷帘门? | 蓝色 | |
| | 5 | 左侧敞开的门口内部可以看到什么设备? | 一台白色的大型机器 | |
|
|
|
|
| ### Example 2 — 4 turns on one image |
|
|
|  |
|
|
| | # | question | answer | |
| |---|---|---| |
| | 1 | 100%感觉下方的英文写了什么? | 100% feeling | |
| | 2 | 白色招牌上最大的红色中文是什么？ | 100%感觉 | |
| | 3 | 门口左侧黑色小牌子上写的门牌号是多少？ | 253 | |
| | 4 | 店铺门口挂着什么颜色的灯笼？ | 红色 | |
|
|
|
|
| ### Example 3 — 4 turns on one image |
|
|
|  |
|
|
| | # | question | answer | |
| |---|---|---| |
| | 1 | which store offers LUXURY AYURVEDA? | SPA CEYLON | |
| | 2 | What is the name of the store with the blue cursive neon sign on the left side of the image? | Sooker | |
| | 3 | What blue neon sign is visible in the background inside the Spa Ceylon store area? | FOREVER NEW | |
| | 4 | What product names are visible on the black shelves in the bottom right foreground? | Daily De-stress and Four Reserve | |
|
|
|
|
| ### Example 4 — 5 turns on one image |
|
|
|  |
|
|
| | # | question | answer | |
| |---|---|---| |
| | 1 | 玻璃门上的门牌号是多少? | 23-3 | |
| | 2 | 招牌左侧的大字店名是什么？ | 怡烤盘 | |
| | 3 | 招牌右侧列出了哪些特色菜品？ | 特色 海鲜 小炒，原味 羊排 牛肉，生蚝 扇贝 虾尾 | |
| | 4 | 门口墙边摆放了几把白色的椅子？ | 4把 | |
| | 5 | 右侧红色招牌上可见的大字是什么？ | 大 | |
|
|
|
|
| ### Example 5 — 4 turns on one image |
|
|
|  |
|
|
| | # | question | answer | |
| |---|---|---| |
| | 1 | Which shop name is written in golden? | shea | |
| | 2 | What slogan is written below the shop name 'shea'? | pamper your body | |
| | 3 | What unit number is visible on the door frame to the left? | B2-47 | |
| | 4 | What large wooden object is displayed in the shop window? | A wagon wheel | |
|
|
|
|
| ### Example 6 — 6 turns on one image |
|
|
|  |
|
|
| | # | question | answer | |
| |---|---|---| |
| | 1 | 这是太和烧肉什么指定放心销售点? | 黄埔 | |
| | 2 | 黄色招牌中间的一行小字写着什么宗旨？ | 以卫生环保安全食品为宗旨、用心打造、做到更好 | |
| | 3 | 招牌上列出了哪些可以批发定购的肉类产品？ | 金牌乳猪、中猪、全猪、烧鹅、烧鸭、叉烧、卤水 | |
| | 4 | 招牌最下方关于老字号的时间描述是什么？ | (老字号)1979年—2016年于名师制作(工厂批发点) | |
| | 5 | 柜台右侧站着的男子穿着什么？ | 红色T恤，系着白色围裙，穿着深色牛仔裤 | |
| | 6 | 柜台左侧站着的女子穿着什么？ | 绿色长袖上衣和蓝色牛仔裤 | |
|
|
|
|
| ### Example 7 — 5 turns on one image |
|
|
|  |
|
|
| | # | question | answer | |
| |---|---|---| |
| | 1 | Who owns the car? | CKERINGTON Central Marching Tigers | |
| | 2 | What text is written in a script font below 'CKERINGTON Central'? | Tigers | |
| | 3 | What text is written in a sans-serif font next to 'Tigers'? | Marching | |
| | 4 | What copyright information is visible on the right side of the trailer? | © 2013 Google | |
| | 5 | What type of vehicle is shown in the image? | A white enclosed trailer | |
|
|
|
|
| ### Example 8 — 5 turns on one image |
|
|
|  |
|
|
| | # | question | answer | |
| |---|---|---| |
| | 1 | 北蔡镇属于哪个区? | 浦东新区 | |
| | 2 | 黑色牌子上最大的字是什么？ | 无烟医疗机构 | |
| | 3 | 黑色牌子上列出了哪三个机构？ | 上海市卫生局、上海市医务工会、上海市爱国卫生运动委员会办公室 | |
| | 4 | 金色牌子下方写着什么？ | 红十字服务总站 | |
| | 5 | 金色牌子中间有什么标志？ | 红十字标志 | |
|
|
|
|
| ### Example 9 — 6 turns on one image |
|
|
|  |
|
|
| | # | question | answer | |
| |---|---|---| |
| | 1 | what is the website of americanfield? | www.americanfield.us | |
| | 2 | What is the main title written in the center seal of the banner? | The ALL-AMERICAN AMERICAN FIELD | |
| | 3 | What specific event or market is described below the main title? | BOS POP-UP MARKET USA | |
| | 4 | Who is credited with making the banner on the right side? | Bullard & Buck | |
| | 5 | What year is listed as 'EST.' inside the central circular seal? | 2012 | |
| | 6 | What is the background design of the banner? | An American flag with stars and stripes | |
|
|
|
|
| ### Example 10 — 4 turns on one image |
|
|
|  |
|
|
| | # | question | answer | |
| |---|---|---| |
| | 1 | 这里是哪里? | 西村新市场 | |
| | 2 | 墙上的大字是什么颜色的？ | 金色 | |
| | 3 | 文字下方安装了一个什么设备？ | 监控摄像头 | |
| | 4 | 墙面上横七竖八的是什么？ | 黑色的电线 | |
|
|
|
|
