Datasets:

lance-format
/

textvqa-lance

license: cc-by-4.0
task_categories:
  - visual-question-answering
  - image-text-to-text
  - image-feature-extraction
language:
  - en
tags:
  - textvqa
  - ocr
  - vqa
  - vision-language
  - lance
  - clip-embeddings
pretty_name: textvqa-lance
size_categories:
  - 10K<n<100K

TextVQA (Lance Format)

Lance-formatted version of TextVQA — VQA where the question requires reading text in the image — sourced from lmms-lab/textvqa.

Each row carries the image bytes, the question, the 10 reference answers, the OCR tokens detected by the dataset's pre-processing, and CLIP image + question embeddings.

Splits

Split	Rows
`validation.lance`	5,000
`train.lance`	34,602

Schema

Column	Type	Notes
`id`	`int64`	Row index within split
`image`	`large_binary`	Inline JPEG bytes
`image_id`	`string?`	TextVQA image id
`question_id`	`string?`	TextVQA question id
`question`	`string`	The question text
`answers`	`list<string>`	10 annotator answers
`answer`	`string`	First answer — used as canonical / FTS target
`ocr_tokens`	`list<string>`	OCR tokens detected on the image
`image_classes`	`list<string>`	OpenImages-style scene tags from the source
`set_name`	`string?`	Source partition (`train`, `val`)
`image_emb`	`fixed_size_list<float32, 512>`	OpenCLIP image embedding (cosine-normalized)
`question_emb`	`fixed_size_list<float32, 512>`	OpenCLIP text embedding of the question

Pre-built indices

IVF_PQ on image_emb and question_emb — metric=cosine
INVERTED (FTS) on question and answer
BTREE on image_id, question_id, set_name

Quick start

import lance
ds = lance.dataset("hf://datasets/lance-format/textvqa-lance/data/validation.lance")
print(ds.count_rows(), ds.schema.names, ds.list_indices())

Cross-modal text→image search

import lance, pyarrow as pa, open_clip, torch

model, _, _ = open_clip.create_model_and_transforms("ViT-B-32", pretrained="laion2b_s34b_b79k")
tokenizer = open_clip.get_tokenizer("ViT-B-32")
model = model.eval().cuda().half()
with torch.no_grad():
    q = model.encode_text(tokenizer(["what brand is on this billboard?"]).cuda())
    q = (q / q.norm(dim=-1, keepdim=True)).float().cpu().numpy()[0]

ds = lance.dataset("hf://datasets/lance-format/textvqa-lance/data/validation.lance")
emb_field = ds.schema.field("image_emb")
hits = ds.scanner(
    nearest={"column": "image_emb", "q": pa.array([q.tolist()], type=emb_field.type)[0], "k": 10},
    columns=["question", "answer", "ocr_tokens"],
).to_table().to_pylist()

Why Lance?

One dataset for images + questions + answers + OCR + dual embeddings + indices — no JSON/feature folders.
Cross-modal search and OCR-text filtering work on the same dataset on the Hub.
Schema evolution: add columns (alternate OCR systems, model predictions) without rewriting the data.

Source & license

Converted from lmms-lab/textvqa. TextVQA is released under CC BY 4.0 by Singh et al. (Facebook AI Research).

Citation

@inproceedings{singh2019towards,
  title={Towards VQA models that can read},
  author={Singh, Amanpreet and Natarajan, Vivek and Shah, Meet and Jiang, Yu and Chen, Xinlei and Batra, Dhruv and Parikh, Devi and Rohrbach, Marcus},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2019}
}