README.md

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

# cute80_cleaned

The **cute80** family of the ElliotVL supervised-fine-tuning pool, **after VLM cleaning**.

| | |
|---|---|
| images | 77 |
| QA turns | 399 |
| answers rewritten by the cleaning pass | 0 |
| QA created by the cleaning pass (`new_qa`) | 359 (90.0%) |
| shards | 1 |

## 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.

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

![case1](samples/case1.jpg)

| # | question | answer |
|---|---|---|
| 1 | What name is printed on the back of the jersey? | RONALDO |
| 2 | What number is displayed on the back of the jersey? | 7 |
| 3 | What color is the player's jersey? | Red |
| 4 | What is the player doing with his arms? | He has his arms outstretched to the sides. |
| 5 | What is visible on the player's left wrist? | A white wristband |


### Example 2 — 6 turns on one image

![case2](samples/case2.jpg)

| # | question | answer |
|---|---|---|
| 1 | Detect all curved text regions in the image. Return a JSON list of objects with key `polygon`, where each polygon is a list of normalized [x, y] points. | [{"polygon":[[0.0977,0.4362],[0.1045,0.4102],[0.1689,0.3659],[0.208,0.3451],[0.2607,0.3216],[0.3555,0.2891],[0.4434,0.2773],[0.5215,0.2708],[0.6309,0.2799],[0.6924,0.2969],[0.7549,0.3203],[0.75,0.3385],[0.6943,0.4323],[0.6045,0.4049],[0.5254,0.3893],[0.4629,0.388],[0.4199,0.3893],[0.3555,0.401],[0.29,0.4167],[0.2412,0.4349],[0.1963,0.4557],[0.1279,0.5026],[0.1074,0.4974]]},{"polygon":[[0.1797,0.6667],[0.2217,0.6849],[0.2676,0.6901],[0.3096,0.6914],[0.3701,0.6953],[0.4688,0.681],[0.5547,0.6628],[0.6133,0.6354],[0.6348,0.7721],[0.5615,0.793],[0.4746,0.8125],[0.4014,0.8216],[0.3301,0.8203],[0.2988,0.8177],[0.251,0.8073],[0.1934,0.793],[0.1553,0.7839]]}] |
| 2 | What text is written on the sign? | SEACREST BEACH |
| 3 | What graphic is depicted in the center of the sign? | A silhouette of a lifeguard tower |
| 4 | What are the colors of the text and the border of the sign? | Dark green |
| 5 | What is the shape of the sign? | Oval |
| 6 | What structure is the sign attached to? | A white wooden structure |


### Example 3 — 8 turns on one image

![case3](samples/case3.jpg)

| # | question | answer |
|---|---|---|
| 1 | Detect all curved text regions in the image. Return a JSON list of objects with key `polygon`, where each polygon is a list of normalized [x, y] points. | [{"polygon":[[0.2197,0.3281],[0.2744,0.2604],[0.4004,0.1797],[0.501,0.1432],[0.5449,0.1549],[0.6592,0.1927],[0.7451,0.25],[0.7959,0.293],[0.8818,0.4076],[0.8281,0.4961],[0.8047,0.4831],[0.7705,0.4336],[0.709,0.3789],[0.6787,0.3477],[0.6172,0.3177],[0.583,0.3047],[0.5146,0.2995],[0.4863,0.3047],[0.3994,0.3542],[0.3486,0.4036],[0.3037,0.4609]]}] |
| 2 | What is the large orange text on the main arch? | BALLY'S |
| 3 | What text is visible on the sign below the main arch? | STATION ENTRANCE |
| 4 | What logo is visible to the left of the 'STATION ENTRANCE' sign? | Starbucks |
| 5 | Describe the entrance structure leading into the building. | A tunnel-like structure formed by white arches illuminated with blue neon lights. |
| 6 | What color are the lights on the large white pillars? | Blue |
| 7 | What is the white object standing on the right side of the image? | A sculpture resembling a hook or 'S' shape on a spherical base. |
| 8 | What is visible in the background behind the arch? | A tall building with many lit windows, likely a hotel or casino tower. |


### Example 4 — 5 turns on one image

![case4](samples/case4.jpg)

| # | question | answer |
|---|---|---|
| 1 | Detect all curved text regions in the image. Return a JSON list of objects with key `polygon`, where each polygon is a list of normalized [x, y] points. | [{"polygon":[[0.2408,0.1699],[0.266,0.1611],[0.288,0.1611],[0.3539,0.1328],[0.423,0.1162],[0.4764,0.0938],[0.5141,0.0957],[0.5644,0.1182],[0.6251,0.1348],[0.6953,0.1611],[0.7204,0.1846],[0.6953,0.2178],[0.6356,0.1826],[0.5874,0.1699],[0.5382,0.1572],[0.4911,0.1514],[0.4031,0.1787],[0.3613,0.1719],[0.3257,0.1885],[0.2921,0.2109],[0.2618,0.2217],[0.244,0.2129]]},{"polygon":[[0.3267,0.7881],[0.3529,0.7871],[0.4157,0.79],[0.5037,0.792],[0.5529,0.8086],[0.5822,0.8027],[0.6199,0.7881],[0.6597,0.7773],[0.6733,0.8193],[0.645,0.8311],[0.5895,0.8496],[0.556,0.8535],[0.5068,0.8613],[0.4304,0.8555],[0.3969,0.8506],[0.3487,0.835],[0.3215,0.8291],[0.3152,0.8125]]}] |
| 2 | What text is embroidered in red along the top curve of the patch? | Carp Team |
| 3 | What text is embroidered in red along the bottom curve of the patch? | eBizU |
| 4 | Describe the central graphic on the patch. | A stylized, smiling face with closed eyes and a black pointed hat (resembling a Daruma doll), with a black carp fish swimming near the bottom right. |
| 5 | What is the small text visible in the bottom right corner of the image? | Powered by DIYTrade.com |


### Example 5 — 5 turns on one image

![case5](samples/case5.jpg)

| # | question | answer |
|---|---|---|
| 1 | What is the brand name written in an arc at the top of the logo on the bottles? | THE DAILY GRIND |
| 2 | What text is printed below the logo on the bottle to the left? | CHILLI |
| 3 | What text is printed below the logo on the bottle to the right? | KETCHUP |
| 4 | What text appears in the lower arc inside the gear logo? | gourmet burgers |
| 5 | What object is depicted in the center of the gear logo? | A burger |


### Example 6 — 5 turns on one image

![case6](samples/case6.jpg)

| # | question | answer |
|---|---|---|
| 1 | Detect all curved text regions in the image. Return a JSON list of objects with key `polygon`, where each polygon is a list of normalized [x, y] points. | [{"polygon":[[0.4112,0.3599],[0.4602,0.3476],[0.4893,0.3497],[0.5294,0.3538],[0.5708,0.364],[0.5938,0.3742],[0.617,0.3844],[0.614,0.4931],[0.5953,0.4869],[0.5953,0.4624],[0.5723,0.4461],[0.5493,0.4359],[0.5187,0.4297],[0.4923,0.4297],[0.4724,0.4252],[0.4464,0.4277],[0.4265,0.4338],[0.4188,0.44],[0.4173,0.4583],[0.405,0.4665]]},{"polygon":[[0.4553,0.4449],[0.5643,0.4632],[0.5613,0.5012],[0.4537,0.4796]]},{"polygon":[[0.4167,0.4734],[0.4252,0.4783],[0.4314,0.4918],[0.4697,0.5041],[0.5098,0.5176],[0.5441,0.5204],[0.572,0.5204],[0.572,0.5094],[0.5971,0.5053],[0.5934,0.5531],[0.5689,0.5645],[0.5395,0.5625],[0.5052,0.5617],[0.4691,0.5531],[0.4415,0.5368],[0.413,0.5184]]}] |
| 2 | What text is printed on the person's blue t-shirt? | IMPERIAL COLLEGE LONDON |
| 3 | What is the person doing with their hands? | Holding the bottom hem of their shirt. |
| 4 | What items are sitting on top of the wooden cabinet to the left? | A potted plant and a red box with floral patterns. |
| 5 | What items are visible on the top shelf of the unit behind the person? | Three white binders or folders. |


### Example 7 — 5 turns on one image

![case7](samples/case7.jpg)

| # | question | answer |
|---|---|---|
| 1 | What text appears at the very top of the wall? | academy entrance |
| 2 | What numbers flank the central blue crest? | 18 and 78 |
| 3 | What Latin phrase is written on the ribbon below the crest? | NIL SATIS NISI OPTIMUM |
| 4 | What large word is displayed at the bottom of the sign? | Everton |
| 5 | What is the boy in the image doing? | He is pointing his finger upwards at the crest. |


### Example 8 — 5 turns on one image

![case8](samples/case8.jpg)

| # | question | answer |
|---|---|---|
| 1 | What name is printed in large white letters across the upper back? | ARSHAVIN |
| 2 | What number is displayed below the name? | 23 |
| 3 | What text is visible at the top center of the collar? | ARSENAL |
| 4 | What are the main colors of the jersey? | Red body with white sleeves |
| 5 | What logo is visible on the right sleeve? | Barclays Premier League logo |


### Example 9 — 4 turns on one image

![case9](samples/case9.jpg)

| # | question | answer |
|---|---|---|
| 1 | What name is printed in white arched letters across the shoulders of the jersey? | BIERHOFF |
| 2 | What large number is printed on the back of the jersey? | 20 |
| 3 | What are the colors of the vertical stripes on the jersey? | Red and black |
| 4 | What text is visible on the CD case located on the shelf in the bottom right background? | MADNESS and It must be love |


### Example 10 — 4 turns on one image

![case10](samples/case10.jpg)

| # | question | answer |
|---|---|---|
| 1 | Detect all curved text regions in the image. Return a JSON list of objects with key `polygon`, where each polygon is a list of normalized [x, y] points. | [{"polygon":[[0.3838,0.2852],[0.4287,0.1615],[0.4482,0.1471],[0.499,0.1328],[0.5586,0.1549],[0.5938,0.1888],[0.6143,0.2865],[0.5664,0.25],[0.5391,0.2461],[0.501,0.2357],[0.4561,0.237]]},{"polygon":[[0.4326,0.75],[0.4727,0.7617],[0.5098,0.7643],[0.5605,0.75],[0.5771,0.7539],[0.5977,0.793],[0.5879,0.819],[0.5176,0.8359],[0.4609,0.8359],[0.4141,0.819],[0.4131,0.7539]]}] |
| 2 | What text is written in the upper and lower sections of the central oval emblem? | "ACM" is at the top and "1899" is at the bottom. |
| 3 | What text is visible on the rectangular patch in the bottom right corner? | "LIGA CALCIO", "XII 07", and "JINJI DESIGN". |
| 4 | Describe the background fabric and the logo in the top left corner. | The fabric has vertical red and black stripes with a perforated mesh texture, and the top left corner features a soccer ball with a green and red ribbon. |