Skip to the content
Know your tech before you buy it. Makers: get listed from $49/mo

MakeSense.ai

makesense.ai

  • Free plan
  • Open source
  • Windows
  • macOS
  • Linux
  • Web
  • Self-hosted
8.3 Club score Visit makesense.ai

What MakeSense.ai is

MakeSense.ai is a browser-based tool for labeling photos and preparing datasets for computer vision projects. It supports point, line, rectangle, polygon, and label annotations, with AI integrations including YOLOv5, SSD pretrained on COCO, and PoseNet. Users can load their own YOLOv5 models exported to TensorFlow.js format. Supported exports include CSV, YOLO, VOC XML, VGG JSON, and COCO JSON; imports include YOLO, VOC XML, and COCO JSON, depending on annotation type. The project says images are not sent to a server and that its AI features run using TensorFlow.js. It works in a browser across operating systems, and the project documents local setup as well as a Docker setup serving the application on localhost:3000. The project describes it as suitable for small computer vision deep-learning projects and dataset preparation. MakeSense.ai was founded in 2019, is licensed under GPL-3.0, and credits Piotr Skalski as author and copyright holder. The tool is free to use.

Who it suits

MakeSense.ai suits people preparing datasets for small computer vision projects who need photo annotation and supported exports. Its browser-based workflow and documented local setup may suit users who prefer to work without sending images to a server.

Good to know, and what to watch

  • Supports five annotation types.
  • Lists YOLOv5, SSD, and PoseNet integrations.
  • Exports to CSV and four dataset formats.
  • Project says images are not sent to a server.
  • Offers documented local and Docker setup.
  • Some format and annotation combinations are not supported.
  • Described as suitable for small projects and dataset preparation.

Free

Free

Free plan

Getting started

  1. Open https://www.makesense.ai/ in a browser
  2. Use the web application on Linux, macOS, Windows, or another operating system
  3. Load images for point, line, rectangle, polygon, or label annotation
  4. Optionally use a listed model integration or load a YOLOv5 TensorFlow.js model
  5. Export the dataset as CSV, YOLO, VOC XML, VGG JSON, or COCO JSON
  6. For local use, follow the documented setup or Docker configuration at localhost:3000

Limits to check

The format matrix marks some annotation and import/export combinations as unsupported. Custom model loading is limited to YOLOv5 models exported for TensorFlow.js.

The verdict

MakeSense.ai covers photo annotation, model-assisted labeling, and dataset export without sending images to a server, according to the project. Check its format matrix to confirm that the needed import and export combinations are supported.

Questions people ask

Is MakeSense.ai free?

Yes. The Free plan is listed as free.

Which platforms can run it?

It runs in a browser across operating systems, with Linux, macOS, Windows, web, and self-hosted use listed.

Which annotation types are available?

Point, line, rectangle, polygon, and label annotations are supported.

Can it use AI models?

Listed integrations include YOLOv5, SSD pretrained on COCO, and PoseNet. Users can load their own YOLOv5 models exported to TensorFlow.js.

What formats can it import and export?

Imports include YOLO, VOC XML, and COCO JSON. Exports include CSV, YOLO, VOC XML, VGG JSON, and COCO JSON, depending on annotation type.

Is it open source?

The project is licensed under GPL-3.0.

What the maker publishes

AI models
Its listed AI integrations include YOLOv5, SSD pretrained on COCO, and PoseNet. source
Annotation features
The tool supports point, line, rectangle, polygon, and label annotations. source
API and SDK documentation
The public documentation presents no product API or SDK; its documented extensibility is local code, model loading and inference-server connection. source
Cross-platform
The project describes itself as cross-platform because it runs in a browser across operating systems. source
Cross-platform use
The project says it works in a browser across operating systems and requires no complicated installation. source
Custom models
Users can load their own YOLOv5 models exported to TensorFlow.js format. source
Docker self-hosting
The repository includes Docker instructions for running the application locally. source
Documentation
The public documentation contains separate pages for imports, exports, shortcuts and local setup. source
Export formats
The export format table lists CSV, YOLO, VOC XML, VGG JSON, and COCO JSON for supported annotation types. source
Export formats
Supported exports include CSV, YOLO, VOC XML, VGG JSON and COCO JSON, depending on annotation type. source
Founded
2019 source
Image classification
Image-level labels are available for classification tags. source
Implementation stack
The application is written in TypeScript and uses React with Redux. source
Import formats
The import format table lists YOLO, VOC XML, and COCO JSON for supported annotation types. source
Import formats
Supported imports include YOLO and COCO JSON for rectangles and COCO JSON for polygons; other combinations are marked unsupported or not applicable. source
Independent praise: privacy
CheckThat.ai characterizes the tool as free, open-source, browser-based and processing data locally for privacy. source
Independent use
Researchers have used MakeSense.ai in scientific image-labeling work, including X-ray scattering datasets. source
Intended users
The project describes the tool as suitable for small computer vision deep learning projects and dataset preparation. source
Latest release
The releases page lists version 1.11.0-alpha dated 20 December 2022 as the latest release shown. source
Licence
The code is released under the GNU GPL-3.0 licence. source
Limit: batch workflows
The same paper reports that MakeSense.ai lacks batch labeling and similarity-based annotation suggestions. source
Limit: image adjustment
The paper reports no built-in brightness or contrast adjustment tools. source
Limit: model accuracy
A 2021 CEUR paper describes the neural-network choices as limited and reports low object-detection accuracy in its comparison. source
Limit: thumbnail information
The paper reports that thumbnails show whether an image is labeled but not its specific labels or label totals. source

Checked against the maker's pages on 21 Oct 2026.