How to annotate video data for object detection with Diffgram

Anthony Sarkis
Oct 4, 2018 · 4 min read

Annotating video data for computer vision can be a challenging task. The Visual Geometry Group at Oxford released their video annotation tool in 2017. It’s a good labeling tool for getting started with computer vision.

One challenge can be collaborating with other people. Most visual intelligence projects are the work of teams of people. For example one person may collect data, while many annotate, and others build the machine learning system.

Here we walk through a practical example of how to annotate a video with Diffgram. Here’s the end result:

1. Create a new project

Data is isolated by project in Diffgram. Create a new project to try out using it.

  • Click “Projects”
  • Click “New”
  • Choose a name
  • Click “Create”
Image for post
Image for post
Image for post

2. Upload a video

For tips on data set collection please see How to create AI ready data for Object Detection.

  • Click “Upload”
  • Drag and drop or click in side the box to use file selector
  • Click Done
  • Wait for images to process

Supported video formats are:

.mp4, .mov, .avi
Image for post
Image for post
Image for post
Image for post

Upload your video. Diffgram supports files up to 1 GB.

Once upload is complete files show up in the Media browser, on the Annotate page.

Large videos may take several minutes to process — click refresh to see changes.

Image for post
Image for post

3. Create a label

In order to map the meaning onto your data you need to create a label. A label represents the meaning behind what is in the data.

Image for post

4. Create a video annotation

Image for post
Video Interface
  • Click on the label “car” to open the Sequence navigator
Image for post
Sequence Navigator
  • Draw a box around all the objects in the image (for example all the cars).
Image for post
  • Tip: Use the “speed” control to slow down playback. This is especially useful for high frame rate videos.
  • Play the video.
  • Stop the video after a few frames.
  • Annotate the objects, using the thumbnail images in the Sequence Navigator to help. For example, the car with id #1 is the same car throughout the whole video.
Image for post
Caught mid-draw
Image for post
Image for post

5. Interpolate

Knowing it’s the same car allows for point to point interpolation between frames. This means you annotate key frames, ie every 5th or 10th frame, and get the annotations in between filled in for you.

Image for post
  • After you are finished annotating click “Interpolate”.
  • Now we can go back to the start of the video and review our work.
Image for post

6. Invite teammates to help you annotate

Annotating often requires subject matter experts, such as Professionals, Engineers and Doctors, etc. You can invite teammates to help you annotate, upload data, and construct machine learning models.

Image for post
  • Click “Share” in top right corner
  • Enter their email and what permission you want
  • Hit invite
Image for post

Your teammate will receive an email inviting them to the project.

If they have a Diffgram account they will be instantly added to the project, else they will receive an email with an invite to create a new account.


That’s the basics of using Diffgram to create video annotations!

If you are interested in participating in the beta signup here.

Thanks for reading!

Diffgram

Supervision for Deep Learning

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch

Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore

Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store