README.md

---
license: other
task_categories:
- visual-question-answering
language:
- en
---

# ctw_cleaned

The **ctw__x** family of the ElliotVL supervised-fine-tuning pool, **after VLM cleaning**.

| | |
|---|---|
| images | 17,650 |
| QA turns | 173,451 |
| answers rewritten by the cleaning pass | 0 |
| QA created by the cleaning pass (`new_qa`) | not measured for this family |
| shards | 152 |

## 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 109 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 — 15 turns on one image

![case1](samples/case1.png)

| # | question | answer |
|---|---|---|
| 1 | What does the text at [[0.07, 0.44], [0.11, 0.44], [0.11, 0.46], [0.07, 0.46]] say? | 明海地产 |
| 2 | What does the text at [[0.11, 0.44], [0.15, 0.44], [0.15, 0.46], [0.11, 0.46]] say? | 易生·印帝安 |
| 3 | What does the text at [[0.94, 0.45], [0.94, 0.45], [0.94, 0.46], [0.94, 0.46]] say? | 成和王 |
| 4 | What does the text at [[0.95, 0.44], [1.0, 0.44], [1.0, 0.46], [0.95, 0.46]] say? | 断桥铝五金 |
| 5 | What does the text at [[0.99, 0.47], [1.0, 0.47], [1.0, 0.48], [0.99, 0.48]] say? | 修电 |
| 6 | What does the text at [[-0.0, 0.43], [0.03, 0.43], [0.03, 0.45], [-0.0, 0.45]] say? | 青缘 |
| 7 | What does the text at [[0.01, 0.5], [0.05, 0.5], [0.05, 0.51], [0.01, 0.51]] say? | 平房四合院租售 |
| 8 | What does the text at [[0.03, 0.51], [0.05, 0.51], [0.05, 0.52], [0.03, 0.52]] say? | 过户咨询 |
| 9 | What does the text at [[0.02, 0.52], [0.05, 0.52], [0.05, 0.52], [0.02, 0.52]] say? | 按揭抵押贷款 |
| 10 | What does the text at [[0.02, 0.52], [0.05, 0.52], [0.05, 0.53], [0.02, 0.53]] say? | 市内拆迁咨询 |
| 11 | What brand of car is the black sedan in the foreground left lane? | Audi |
| 12 | What brand of car is the black sedan in the right lane? | Mercedes-Benz |
| 13 | What color is the minivan driving behind the black Audi? | Orange |
| 14 | What feature separates the two directions of traffic? | A median strip with green bushes and a low green fence |
| 15 | What color is the taxi visible further down the road? | Yellow |


### Example 2 — 5 turns on one image

![case2](samples/case2.png)

| # | question | answer |
|---|---|---|
| 1 | What does the text at [[0.13, 0.45], [0.19, 0.45], [0.19, 0.46], [0.13, 0.46]] say? | 杭州大厦 |
| 2 | What does the text at [[0.21, 0.44], [0.28, 0.44], [0.28, 0.46], [0.21, 0.46]] say? | 城市生活馆 |
| 3 | What does the text at [[0.3, 0.44], [0.33, 0.44], [0.33, 0.45], [0.3, 0.45]] say? | 武林府店 |
| 4 | What text is visible on the storefront sign on the ground floor of the tall building on the right? | FlyApp |
| 5 | What is the color of the horizontal railing running along the top of the concrete barrier in the foreground? | Red |


### Example 3 — 12 turns on one image

![case3](samples/case3.png)

| # | question | answer |
|---|---|---|
| 1 | What does the text at [[0.52, 0.44], [0.53, 0.44], [0.53, 0.45], [0.52, 0.45]] say? | 西四环 |
| 2 | What does the text at [[0.54, 0.45], [0.55, 0.45], [0.55, 0.45], [0.54, 0.45]] say? | 四环路 |
| 3 | What does the text at [[0.79, 0.43], [0.83, 0.43], [0.83, 0.46], [0.79, 0.46]] say? | 名师状元 |
| 4 | What does the text at [[0.91, 0.42], [0.98, 0.42], [0.98, 0.45], [0.91, 0.45]] say? | 派乐汉堡 |
| 5 | What does the text at [[0.93, 0.28], [0.96, 0.28], [0.96, 0.3], [0.93, 0.3]] say? | 信德留学 |
| 6 | What does the text at [[0.97, 0.27], [1.0, 0.27], [1.0, 0.3], [0.97, 0.3]] say? | 做中 |
| 7 | What does the text at [[0.09, 0.49], [0.11, 0.49], [0.11, 0.5], [0.09, 0.5]] say? | 西直门 |
| 8 | What does the text at [[0.04, 0.49], [0.06, 0.49], [0.06, 0.5], [0.04, 0.5]] say? | 香山 |
| 9 | What number is displayed on the LED screen of the bus? | 360 |
| 10 | What text is written on the red hood of the car in the foreground? | 腾讯街景地图 |
| 11 | What white geometric shape is painted on the road surface in the center lane? | A diamond |
| 12 | What object separates the bus lane from the main traffic lanes? | A white metal fence |


### Example 4 — 13 turns on one image

![case4](samples/case4.png)

| # | question | answer |
|---|---|---|
| 1 | What does the text at [[0.25, 0.4], [0.31, 0.4], [0.31, 0.44], [0.25, 0.44]] say? | 雨林 |
| 2 | What does the text at [[0.3, 0.44], [0.35, 0.44], [0.35, 0.46], [0.3, 0.46]] say? | 专业烫染 |
| 3 | What does the text at [[0.23, 0.5], [0.25, 0.5], [0.25, 0.51], [0.23, 0.51]] say? | 美甲 |
| 4 | What does the text at [[0.36, 0.44], [0.38, 0.44], [0.38, 0.51], [0.36, 0.51]] say? | 友谊综合楼 |
| 5 | What does the text at [[0.41, 0.47], [0.44, 0.47], [0.44, 0.47], [0.41, 0.47]] say? | 友谊综合 |
| 6 | What does the text at [[0.63, 0.44], [0.65, 0.44], [0.65, 0.45], [0.63, 0.45]] say? | 达兰 |
| 7 | What does the text at [[0.65, 0.44], [0.66, 0.44], [0.66, 0.45], [0.65, 0.45]] say? | 服饰 |
| 8 | What does the text at [[0.91, 0.41], [0.96, 0.41], [0.96, 0.43], [0.91, 0.43]] say? | 中国邮政 |
| 9 | What does the text at [[0.79, 0.44], [1.0, 0.44], [1.0, 0.46], [0.79, 0.46]] say? | 中奖、银行卡升级等名义要求将款转入指定账户 |
| 10 | What English text appears below the red Chinese characters '雨林'? | YULIN |
| 11 | What English text appears below the green '中国邮政' sign? | CHINA POST |
| 12 | What are the blue house numbers visible on the building pillars? | 74 and 72 |
| 13 | What is the phone number printed on the side of the red and silver taxi? | Tel: 96529 |


### Example 5 — 10 turns on one image

![case5](samples/case5.png)

| # | question | answer |
|---|---|---|
| 1 | What does the text at [[0.81, 0.5], [0.85, 0.5], [0.85, 0.51], [0.81, 0.51]] say? | 滨盛路 |
| 2 | What does the text at [[0.55, 0.47], [0.57, 0.47], [0.57, 0.48], [0.55, 0.48]] say? | 药房 |
| 3 | What does the text at [[0.95, 0.46], [1.0, 0.46], [1.0, 0.47], [0.95, 0.47]] say? | 生鲜超市 |
| 4 | What does the text at [[0.67, 0.48], [0.71, 0.48], [0.71, 0.49], [0.67, 0.49]] say? | 树立正确理财观念 |
| 5 | What does the text at [[0.78, 0.35], [0.82, 0.35], [0.82, 0.37], [0.78, 0.37]] say? | 银行 BANK OF CHINA |
| 6 | What does the text at [[0.12, 0.45], [0.15, 0.45], [0.15, 0.46], [0.12, 0.46]] say? | XX小学 |
| 7 | What English text is written on the large white sign above the bank entrance? | BANK OF CHINA |
| 8 | What red text is visible on top of the tall building in the background? | SOLAR |
| 9 | What number is visible on the small sign near the bank entrance? | 24 |
| 10 | What is the person on the left side of the image doing? | Crossing the street at a crosswalk |


### Example 6 — 8 turns on one image

![case6](samples/case6.png)

| # | question | answer |
|---|---|---|
| 1 | What does the text at [[0.37, 0.46], [0.38, 0.46], [0.38, 0.47], [0.37, 0.47]] say? | 生 |
| 2 | What does the text at [[0.42, 0.45], [0.43, 0.45], [0.43, 0.47], [0.42, 0.47]] say? | 养发 |
| 3 | What does the text at [[0.49, 0.45], [0.51, 0.45], [0.51, 0.47], [0.49, 0.47]] say? | 美容 |
| 4 | What does the text at [[0.53, 0.45], [0.54, 0.45], [0.54, 0.47], [0.53, 0.47]] say? | 养发 |
| 5 | What are the makes of the two fully visible cars parked in the foreground? | A grey Honda and a white Volkswagen. |
| 6 | What symbol is visible at the top of the blue sign on the left? | A white 'P' symbol. |
| 7 | What structure separates the sidewalk from the building's ground floor area? | A black metal fence. |
| 8 | What is the color of the windows on the upper floors of the building? | Green. |


### Example 7 — 5 turns on one image

![case7](samples/case7.png)

| # | question | answer |
|---|---|---|
| 1 | What does the text at [[0.49, 0.06], [0.59, 0.06], [0.59, 0.09], [0.49, 0.09]] say? | 保亿创艺 |
| 2 | What does the text at [[0.49, 0.41], [0.5, 0.41], [0.5, 0.42], [0.49, 0.42]] say? | 金 |
| 3 | What text is visible on the top of the building on the right side? | BEING MATE |
| 4 | How many people are walking on the sidewalk in the lower left? | Two |
| 5 | What are the young trees in the grassy area supported by? | Wooden stakes |


### Example 8 — 16 turns on one image

![case8](samples/case8.png)

| # | question | answer |
|---|---|---|
| 1 | What does the text at [[0.03, 0.48], [0.04, 0.48], [0.04, 0.49], [0.03, 0.49]] say? | 装修 |
| 2 | What does the text at [[0.04, 0.48], [0.07, 0.48], [0.07, 0.5], [0.04, 0.5]] say? | 瑞通装饰 |
| 3 | What does the text at [[0.38, 0.39], [0.42, 0.39], [0.42, 0.41], [0.38, 0.41]] say? | 高德 |
| 4 | What does the text at [[0.17, 0.47], [0.18, 0.47], [0.18, 0.48], [0.17, 0.48]] say? | 旺香茶莊 |
| 5 | What does the text at [[0.19, 0.47], [0.21, 0.47], [0.21, 0.48], [0.19, 0.48]] say? | 烟酒销售 |
| 6 | What does the text at [[0.87, 0.43], [0.97, 0.43], [0.97, 0.45], [0.87, 0.45]] say? | 山晋云中刀削面 |
| 7 | What does the text at [[0.87, 0.37], [0.99, 0.37], [0.99, 0.43], [0.87, 0.43]] say? | 福美家生 |
| 8 | What does the text at [[0.85, 0.45], [0.87, 0.45], [0.87, 0.46], [0.85, 0.46]] say? | 北京农商 |
| 9 | What does the text at [[0.98, 0.42], [0.99, 0.42], [0.99, 0.44], [0.98, 0.44]] say? | 二楼 |
| 10 | What does the text at [[0.15, 0.46], [0.16, 0.46], [0.16, 0.48], [0.15, 0.48]] say? | 洗衣 |
| 11 | What does the text at [[0.76, 0.47], [0.77, 0.47], [0.77, 0.48], [0.76, 0.48]] say? | 牙牙 |
| 12 | What is visible at the very bottom center of the image? | The blue hood of the vehicle from which the photo is taken. |
| 13 | Describe the tall building on the right side of the street. | A tall, light green residential building with many windows and air conditioning units. |
| 14 | What is the prominent structure on the far left side of the image? | A concrete utility pole with a large tangle of black wires. |
| 15 | What vehicle is driving in the middle of the road? | A white car (likely a minivan) driving away from the camera. |
| 16 | What separates the lanes of traffic? | A median strip covered with low bushes and green fencing. |


### Example 9 — 12 turns on one image

![case9](samples/case9.png)

| # | question | answer |
|---|---|---|
| 1 | What does the text at [[0.0, 0.42], [0.02, 0.42], [0.02, 0.43], [0.0, 0.43]] say? | 六空间 |
| 2 | What does the text at [[-0.0, 0.38], [0.04, 0.38], [0.04, 0.41], [-0.0, 0.41]] say? | 中南国际商城 |
| 3 | What does the text at [[0.06, 0.41], [0.07, 0.41], [0.07, 0.42], [0.06, 0.42]] say? | 建材 |
| 4 | What does the text at [[0.06, 0.42], [0.07, 0.42], [0.07, 0.43], [0.06, 0.43]] say? | 家具 |
| 5 | What does the text at [[0.04, 0.43], [0.05, 0.43], [0.05, 0.44], [0.04, 0.44]] say? | 西溪 |
| 6 | What does the text at [[0.03, 0.43], [0.06, 0.43], [0.06, 0.44], [0.03, 0.44]] say? | 国家湿地公园 |
| 7 | What does the text at [[0.11, 0.01], [0.19, 0.01], [0.19, 0.07], [0.11, 0.07]] say? | 中南国际大厦 |
| 8 | What does the text at [[0.25, 0.03], [0.32, 0.03], [0.32, 0.08], [0.25, 0.08]] say? | 中南国际大厦 |
| 9 | What is the color scheme of the fence in the foreground? | White and blue |
| 10 | What type of vehicle is driving in the center lane? | A teal and silver taxi |
| 11 | What large vehicle is visible on the left side of the road? | A beige bus |
| 12 | What is visible in the background on the right side? | Buildings under construction with cranes |


### Example 10 — 5 turns on one image

![case10](samples/case10.png)

| # | question | answer |
|---|---|---|
| 1 | What does the text at [[0.27, 0.47], [0.43, 0.47], [0.43, 0.48], [0.27, 0.48]] say? | 湘湖美食广场 |
| 2 | What does the text at [[0.17, 0.44], [0.18, 0.44], [0.18, 0.45], [0.17, 0.45]] say? | 头 |
| 3 | What white structure runs along the bottom of the image? | A white metal fence. |
| 4 | What are the thin vertical objects in the landscaped area supported by? | Wooden stakes. |
| 5 | What natural feature is visible in the distance behind the building? | Mountains. |