---
license: cc-by-nc-4.0
configs:
- config_name: train
  data_files:
  - split: train
    path: "data/train/*.parquet"
- config_name: test
  data_files:
  - split: test
    path: "data/test/*.parquet"
- config_name: validation
  data_files:
  - split: validation
    path: "data/val/*.parquet"
- config_name: default
  data_files:
  - split: train
    path: "data/train/*.parquet"
  - split: test
    path: "data/test/*.parquet"
  - split: validation
    path: "data/val/*.parquet"
task_categories:
- image-text-to-text
language:
- en
size_categories:
- 100K<n<1M
dataset_info:
- config_name: train
  features:
  - name: images
    list:
      image:
        decode: true
  - name: id
    dtype: string
  - name: messages
    list:
    - name: role
      dtype: string
    - name: content
      list:
      - name: type
        dtype: string
      - name: text
        dtype: string
  - name: origin_dataset
    dtype: string
  - name: raw_metadata
    dtype: string
  splits:
  - name: train
    num_bytes: 47646918572
    num_examples: 535881
- config_name: test
  features:
  - name: images
    list:
      image:
        decode: true
  - name: id
    dtype: string
  - name: messages
    list:
    - name: role
      dtype: string
    - name: content
      list:
      - name: type
        dtype: string
      - name: text
        dtype: string
  - name: origin_dataset
    dtype: string
  - name: raw_metadata
    dtype: string
  splits:
  - name: test
    num_bytes: 13564137567
    num_examples: 152921
- config_name: validation
  features:
  - name: images
    list:
      image:
        decode: true
  - name: id
    dtype: string
  - name: messages
    list:
    - name: role
      dtype: string
    - name: content
      list:
      - name: type
        dtype: string
      - name: text
        dtype: string
  - name: origin_dataset
    dtype: string
  - name: raw_metadata
    dtype: string
  splits:
  - name: validation
    num_bytes: 6913653907
    num_examples: 76384
- config_name: default
  features:
  - name: images
    list:
      image:
        decode: true
  - name: id
    dtype: string
  - name: messages
    list:
    - name: role
      dtype: string
    - name: content
      list:
      - name: type
        dtype: string
      - name: text
        dtype: string
  - name: origin_dataset
    dtype: string
  - name: raw_metadata
    dtype: string
  splits:
  - name: train
    num_bytes: 47646918572
    num_examples: 535881
  - name: test
    num_bytes: 13564137567
    num_examples: 152921
  - name: validation
    num_bytes: 6913653907
    num_examples: 76384
  download_size: 67867722531
  dataset_size: 68124710046
---

# PlantExpertVQA

PlantExpertVQA is a large-scale visual question answering (VQA) dataset built to advance
vision-language models for agricultural decision-making and interactive plant disease
diagnosis. It is compiled from **45 open-source datasets** (including the widely-used
PlantVillage corpus) and comprises **765,186 expert-verified question-answer pairs**
grounded over **150,841 images**, spanning **38 crop species** and **89 disease
conditions**.

This dataset is indexed on https://project-agml.github.io/ as part of the AgML python
library. Standardized to the HF `image_text_to_text` format with a single conversational
`messages` schema, converted to **Parquet** with image bytes embedded directly.

## Dataset Construction

Questions are organized into **9 distinct categories** spanning **3 levels of cognitive
complexity**, generated via a two-stage pipeline (template-based QA synthesis from image
metadata, followed by multi-stage linguistic re-engineering) and iteratively validated by
domain botanists for scientific accuracy.

| Cognitive Level | Question Categories |
|---|---|
| **Level 1 — Foundational Perception & Identification** | Existence & Sanity Check, Plant Species Identification, General Health Assessment |
| **Level 2 — Detailed Analysis & Verification** | Visual Attribute Grounding, Detailed Verification |
| **Level 3 — Higher-Order Reasoning & Inference** | Specific Disease Identification, Comprehensive Description, Causal Reasoning, Counterfactual Reasoning |

The dataset also underwent linguistic diversification (359% question vocabulary growth
via expert-validated paraphrasing), structural rebalancing of skewed binary-answer
categories, and a two-phase automated + expert quality review pipeline (vagueness scoring,
semantic dissonance detection, relative simplicity checks) to remove low-value or
non-verifiable QA pairs.

## Splits

Each split is hosted as its **own independent config**, so downloading one split never
triggers a download of the others.

| Config | Split | Rows | Size |
|---|---|---|---|
| `train` | `train` | 535,881 | ~47.6 GB |
| `test` | `test` | 152,921 | ~13.6 GB |
| `validation` | `validation` | 76,384 | ~6.9 GB |
| `default` | all three | 765,186 | ~67.9 GB |

## Usage

```python
from datasets import load_dataset

# Load ONLY the train split — isolated download, no test/validation fetched
ds = load_dataset("Project-AgML/PlantExpertVQA", "train")

# Load ONLY the test split
ds = load_dataset("Project-AgML/PlantExpertVQA", "test")

# Load ONLY the validation split
ds = load_dataset("Project-AgML/PlantExpertVQA", "validation")

# Load everything (all three splits combined under one DatasetDict)
ds = load_dataset("Project-AgML/PlantExpertVQA")   # equivalent to "default"

first = ds["train"][0] if "train" in ds else ds[0]

# Access an image — decoded to PIL automatically
img = first["images"][0]
img.show()
```

## Schema

Every record shares the SAME columns so heterogeneous AgML datasets concatenate cleanly:
`id`, `images` (embedded image bytes), `messages`, `origin_dataset`, and `raw_metadata`.

`raw_metadata` is a JSON-encoded string holding source fields not folded into `messages`
(here: `file_names` pointing to the original image path, `qa_id`, `image_id`, `crop`,
`disease`, `category`, `severity`, `answer_type`, `question_category`, `cognitive_level`,
and `dataset_source` — identifying which of the 45 compiled source datasets each QA pair
originated from); restore it with `json.loads(row["raw_metadata"])`. Image placeholders in
`messages` align 1:1 with the `images` column.

## Citation

```bibtex
@article{sakib2025plantexpertvqa,
      title={PlantExpertVQA: A Visual Question Answering Dataset for Benchmarking Vision-Language Models in Plant Science},
      author={Sakib, Syed Nazmus and Haque, Nafiul and Hossain, Mohammad Zabed and Arman, Shifat E.},
      year={2025},
      eprint={2508.17117},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2508.17117}
}

Sakib, Syed Nazmus; Haque, Nafiul; Hossain, Mohammad Zabed; Arman, Shifat E. (2025), "PlantExpertVQA: A Visual Question Answering Dataset for Benchmarking Vision-Language Models in Plant Science", arXiv:2508.17117
```

## License

Released under **CC BY-NC 4.0**. Compiled from 45 open-source datasets, each retaining
attribution to its original source per the authors' documentation. This license
information is for reference only and does not constitute legal advice — refer to the
original repository for authoritative license terms:
https://huggingface.co/datasets/SyedNazmusSakib/PlantExpertVQA.