What is the Annotation Tool Annofab?

1. What is Annofab?

1-1. What is Annofab?

The cloud-based all-in-one annotation tool Annofab is a data creation tool developed in 2018 by Kurusugawa Densan, a venture company originating from Nagoya Institute of Technology, which is headquartered in Nagoya.

By registering an account, you can use a wide range of features for this cloud-based web tool for free, allowing you to use it immediately in your browser without being significantly affected by the PC specifications or environment of the annotator.

Since the release in 2018, continuous improvements and bug fixes have been made to the tool, making it a more reliable and trustworthy option.

>>Training Data Creation Service Annofab

2. Benefits of Annofab

2-1. Wide Compatibility and Versatile Label and Attribute Settings

 

Annofab currently supports image and video annotation.

Image annotation

• Image classification (classification) for labeling the entire image
• Region extraction that surrounds a specific area in the image (Semantic Segmentation and Instance Segmentation)
• Object detection (object detection) that labels specific objects in images with rectangles or polygons
• Continuous lines (polylines) suitable for linear objects such as vehicle lanes and centerlines on roads
• Annotation indicating position at a single point (Point)

 

Video annotation

• Interval annotation on the timeline to capture temporal changes such as actions and blinking
• Classification Annotation of Entire Video Data

 

You can also assign various information that can be read from the subject as attributes.

• Tracking ID (an ID assigned to indicate that it is the same object before and after consecutive images)
• Numeric Input
• Radio Button
• Dropdown
• Free description (can also accommodate transcription of text) etc.

 

With such a variety of configuration options, it is possible to accommodate various annotation projects.

In addition, various settings can be managed collectively on the 'Annotation Specification' screen, and the settings can be exported as JSON data, allowing for easy reuse in other projects, demonstrating its versatility.

 

Furthermore, support for spatial axis annotation for three-dimensional data is planned for the future. This will enable the capture and tracking of details (such as posture, shape, and condition) of objects and backgrounds captured by LiDAR.

2-2. Improving Work Efficiency through Flexible Shortcut Settings

Annofab offers customizable shortcuts for improving work efficiency. While common tools provide little flexibility with only the default operations set, Annofab allows you to freely set shortcuts along with class and attribute settings for each project.

As a result, you can focus on annotation work with minimal mouse operations.

In addition to shortcuts, the UI elements on the screen can be freely rearranged, allowing each annotator to create a more user-friendly workspace.

Annofab can be said to be a tool that is well-considered for improving the efficiency of annotation work that inevitably requires manual labor.

2-3. Various data that provides an overview of the project

 

Comprehensive statistical data that is essential for AI developers and project managers to grasp the entire project is available.

 

• Overall Project Progress
• Cumulative Work Hours by Phase
• Work Hours by Member
• Average working time per image
• Number of Inspection Remarks
• Number of tasks returned by member
• Number of Annotations by Label
• List of annotation data search results by label, etc.

 

You can also manage the progress of the smallest unit of annotation work, called a "task," in a list format. It is possible to know who is working on it, when, how far they have progressed in each phase, what the current status is, and how much time has been spent on each phase.

Annotators can easily display only the work tasks they have completed with the push of a button, making it easy to check the remaining tasks they have.

By easily grasping such information, it is possible to reduce the effort and man-hours required to manage annotation projects.

2-4. Smooth Communication with QA Checkers

In annotation work, the checking tasks performed by checkers and acceptance are very important processes. Annofab has features that point out annotation mistakes and convey comments. To ensure that all interactions between annotators and checkers can be completed smoothly within Annofab, clarity and work efficiency are taken into account, from the way comments are displayed to responses.

For frequently occurring comments, you can use the comment support feature called "standard comments" to prevent discrepancies in the points raised by multiple checkers and to avoid misunderstandings in communication.

2-5. Highly Flexible Annotation Workflow Settings

 

In Annofab's project settings, you can set the number of inspections and the inspection rate to any desired value in the phase after the annotation work. This feature allows for the design of an annotation workflow that meets the required quality. Since you can create as many projects as you like, you can create multiple test projects during the trial phase to refine the most optimal workflow and specifications.

It is also possible to set the inspection rate for each individual annotator as desired, allowing for configurations based on annotation proficiency. For example, while the overall project may have a 70% acceptance inspection rate, it is possible to set a 100% rate specifically for person A, who joined the project midway, enabling a design that considers thorough inspection and feedback for training new employees.

2-6. Advantages of Cloud-Based Tools

Annofab is a cloud-based annotation tool, so its appeal lies in the speed of implementation, as it can be used immediately with just user registration. Since the work data is exchanged over encrypted HTTPS communication with the server where Annofab is hosted, there is no need to manage it on local machines such as general annotators' PCs.

Additionally, annotation projects and data can be configured to restrict access to data outside of each annotator's assigned scope by specifying user permissions for each project.

Additionally, even during the project, you can share progress and annotated data with the client without hindering the work of the annotators.

2-7. Other Benefits

It includes various features such as settings for how to assign the smallest unit of work tasks to annotators (random, selection, task properties), the ability to insert images and tables, formatting options, and the publication function of the annotation work guide equipped with many editor features. It also supports integration with external systems through Web API, Webhook, and CLI. Additionally, there are comprehensive documents such as user manuals that make extensive use of GIF images.

>>User Manual (Annofab)

3. Disadvantages of Annofab

3-1. Disadvantages of Online Tools

Annofab is an online annotation tool that operates on the web. Therefore, there are cases where the speed of loading the next image within a task or actions such as loading video tasks depend on the individual annotator's network environment speed and PC specifications.

Additionally, there may be rare cases where issues occur with Annofab itself due to the influence of server or tool systems. If any phenomena that hinder usage occur, we handle them through the incident reporting form.

>>How should I report behavior that seems to be a bug?

3-2. In the free plan, you cannot choose the data server

Annofab is provided on AWS (Amazon Web Services) managed by its developer, Kurusugawa Densan. Therefore, when using the free plan, the annotation data will also be stored on the same AWS where Annofab exists. If you need to connect to cloud servers other than this AWS (such as your managed AWS, Azure, Google Cloud Storage, etc.) or your own servers, or if you need to handle more confidential data that cannot be taken outside the company, you can use the business plan (subject to review) to support private storage features and firewall features.

What is the difference between the free plan and the business plan?

3-3. Output data is only in JSON

Unlike some annotation tools that support various file output formats, Annofab only outputs fixed-structure JSON data. Therefore, if you need data formats other than JSON, such as CSV, PascalVOC, or YOLO, you will need to prepare a converter or similar tool to change the file format yourself.

3-4. The High Bar Due to Multifunctionality

There are situations where pre-configuration of the tool is necessary to utilize many of the features available in Annofab. For example, when generating tasks, it is possible to set the order and number of images as desired, but it requires a description in a CSV file that conforms to the specified format. Additionally, there are many items to configure for the annotation project, and it may take some time to become familiar with the tool, as the steps and methods for setting the necessary project specifications can be difficult to grasp.

For annotators who actually perform the work, shortcut operations offer many benefits, but some workers may postpone learning them until they feel the need for improved work efficiency.

3-5. Impressions Regarding UI (Screen Design)

Annofab's UI adopts a dark mode UI. It is said that the dark mode UI has features such as 'reducing eye strain and enhancing concentration', but depending on the displayed area, there may be a lot of text information, which could make it difficult to read.

Additionally, from the author's personal impression, the specifications for the UI color settings give a somewhat cluttered feeling across the entire tool, and in terms of the outline display of annotations, it appears to be more jerky compared to other annotation tools, lacking smoothness in display.

4. Use Cases of Annofab

4-1. Case Study 1: Autonomous Driving AI Accuracy Improvement Project / AI Technology Development Manufacturer

 

Work Content
This is an annotation for improving autonomous driving technology, where tagging is performed on various objects and areas using footage and images from a dashcam as material.

・Points for Utilizing Annofab
In long-term projects, a key point is how to minimize the effort of annotators during their work. One of the attractive features of Annofab that shines in this regard is its extensive support for shortcuts. Annotation work can be physically demanding, especially when done for long hours, putting a strain on the annotator's arms. To alleviate such burdens in the field, Annofab was the optimal choice as it allows for efficient operation while performing tasks.

4-2. Case Study 2: Project for Improving OCR Text Recognition Accuracy / AI Technology Development Manufacturer

 

Work Details
This task involves extracting specific text areas from English document images and converting them into data that AI can understand, in order to improve OCR recognition accuracy. The worker is required to manually select specific text areas and add the correct information to each one.

・Points for Utilizing Annofab
Even when utilizing personnel who do not understand Japanese, Annofab allows for an immediate switch of the display language to English through individual settings. The basic annotation operations are easy, and it is possible to respond to comments from checkers instantly, enabling smooth progress in work while effectively using the tool without scattering information.

 

Our site also features numerous case studies from other companies. If you are interested, please refer to this link.

>>AI Annotation Case Studies | Human Science Co., Ltd.

5. Summary

5-1. Summary of the Annotation Tool Annofab

This concludes the introduction of the annotation tool, Annofab, implemented by Human Science.

Among cloud-based annotation tools, Annofab is a highly attractive option as it offers a wide range of features for free while incorporating feedback from users and undergoing numerous improvements. For businesses looking to efficiently manage an annotation tool that can handle images and videos on a limited budget, it may be the ideal solution.

6. Consultations on AI utilization are available at Human Science

6-1. Achieved 48 million teacher data creations

I want to implement AI, but I don't know where to start.

I don't know what to request even if I want to outsource.

In such cases, please feel free to consult Human Science.

At Human Science, we participate in AI development projects across various industries, including natural language processing, medical support, automotive, IT, manufacturing, and construction. To date, we have provided over 48 million high-quality training data through direct transactions with many companies, including GAFAM. We handle a wide range of annotation projects, from small-scale projects to long-term large-scale projects with 150 annotators, regardless of the industry.

>>Human Science Annotation Services

6-2. Resource Management Without Using Crowdsourcing

At Human Science, we do not use crowdsourcing; instead, we proceed with projects using personnel directly contracted by our company. We form teams that can deliver maximum performance based on a solid understanding of each member's practical experience and their evaluations from previous projects.

6-3. Complete Security Room within the Company

At Human Science, we have a security room that meets ISMS standards within our Shinjuku office. Even for highly confidential projects, we provide on-site support. We consider the assurance of confidentiality to be extremely important for any project. Our staff undergoes continuous security training, and we exercise utmost caution in handling information and data, even for remote projects.