Skip to data content

Avlys OTS Data / Image data

Image training data for computer vision

Avlys Data helps computer-vision teams select and prepare image collections for classification, image captioning, visual search and vision-language evaluation. Explore the image types below, then scope the labels, resolution and coverage your task needs.

Request sample & pricingExplore collections

From Hyderabad and Bengaluru, India. Working with teams globally. Availability confirmed for your brief.

Explore the specifications

Image data collections

Compare 6 options by task, coverage and format. Request sample availability and a quote for your chosen scope.

Photography + detailed captions

Avlys OTS Data

Detailed image captions

Real photographs paired with detailed descriptions for testing how precisely vision-language systems understand a scene.

Reference collection scope15,000 captioned images

Objects, spatial relationships and scene details

  • Image captioning
  • Vision-language evaluation
Dataset specifications for Detailed image captions
Formats & labels
JPEG images; JSONL descriptions and metadata. Detailed image descriptions and published split identifiers.
Fit & scope
Natural photographs rather than infographics. Keep training and evaluation splits separate.
Request a quote by email: Detailed image captions

Satellite imagery

Avlys OTS Data

Satellite land-cover images

Labelled Sentinel-2 patches for land-use and land-cover classification, with RGB and multispectral versions.

Reference collection scope27,000 images · 10 classes

13 spectral bands in the multispectral release

  • Land-cover classification
  • Remote sensing
Dataset specifications for Satellite land-cover images
Formats & labels
RGB image and multispectral archives. Land-cover class labels and georeferenced imagery.
Fit & scope
Satellite patches, not street-level images or a current global map.
Request a quote by email: Satellite land-cover images

Apparel classification

Avlys OTS Data

Apparel classification images

Small grayscale clothing images with category labels for lightweight vision baselines and model comparisons.

Reference collection scope70,000 images · 10 classes

28 × 28 pixels · 60,000 train / 10,000 test

  • Image classification
  • Baseline benchmarking
Dataset specifications for Apparel classification images
Formats & labels
Compressed IDX image and label files. One apparel class per image.
Fit & scope
A low-resolution benchmark, not full-resolution product photography or object-detection data.
Request a quote by email: Apparel classification images

Synthetic visual reasoning

Avlys OTS Data

Synthetic visual reasoning

Rendered arrangements of geometric objects with questions, answers and structured scene information.

Reference collection scopeSynthetic scene benchmark

Shape, colour, material, count and spatial relations

  • Visual question answering
  • Compositional reasoning
Dataset specifications for Synthetic visual reasoning
Formats & labels
Rendered images; JSON questions and scene information. Question–answer pairs; scene graphs where supplied in the release.
Fit & scope
Rendered scenes, not natural photographs or specialist diagrams.
Request a quote by email: Synthetic visual reasoning

Historical character recognition

Avlys OTS Data

Historical Japanese characters

Cropped historical Japanese characters for recognition experiments, based on digitised books.

Reference collection scope70,000 images · 10 classes

28 × 28 grayscale Kuzushiji characters

  • Character recognition
  • Historical OCR baselines
Dataset specifications for Historical Japanese characters
Formats & labels
MNIST-format and NumPy archives. Character class labels and a class mapping.
Fit & scope
Isolated characters, not full-page documents. The stated scope covers this reference collection only.
Request a quote by email: Historical Japanese characters

Museum & natural-history imagery

Avlys OTS Data

Cultural & scientific imagery

Art, cultural objects and natural-history imagery with collection metadata for visual search and research.

Reference collection scopeCollection, not a fixed training set

Art, cultural objects and natural-history material

  • Collection search
  • Visual heritage research
Dataset specifications for Cultural & scientific imagery
Formats & labels
Item-specific media and collection metadata. Museum descriptions; task-specific labels would need to be scoped.
Fit & scope
An item-level selection, not a fixed training set. Image resolution, metadata and task-specific annotation needs vary.
Request a quote by email: Cultural & scientific imagery

Build a useful first batch

What to include
in your brief.

Start with the task your model needs to perform. A scoped pilot establishes the selection, review process and delivery expectations before you scale.

See the delivery checklist
  1. Image type, domain and target classes
  2. Resolution, quantity and capture variation
  3. Captions, class labels or other annotation
  4. Training and evaluation splits, including duplicate checks

Before you choose

Image data questions.

Pricing, quality & delivery FAQs
What image data can I request from Avlys?

The catalogue describes captioned photographs, satellite image patches, apparel images, synthetic visual-reasoning scenes, historical characters and cultural or scientific imagery. Their formats and coverage serve different tasks. Availability and the final selection are confirmed for your brief before delivery.

Can Avlys prepare custom image annotations?

Image annotation is scoped around the labels, schema and review process your model needs. A pilot establishes examples, exclusions and acceptance criteria. Custom capture or additional labels are agreed separately from the reference collection described in the catalogue.

From selection to scope

Let’s scope your
image data.

Share the task, coverage, approximate volume and timeline. We’ll discuss availability, sample arrangements and a quote for the agreed work.

Request sample & pricingDiscuss on WhatsApp ↗
sales@avlysai.com

← All Avlys Data services