Background Remover is now updated to version 1.1

Mikhail Tiukavkin
Deelvin Machine Learning
3 min readNov 10, 2020

What is BG Remover

We are pleased to introduce an updated version of the Background Remover, which allows one to change the background on any video with a person in it. This service could be useful for those who start filming their video blog and do not have a green screen or technical capabilities to implement chroma key technology. This technology can also be used for live-broadcasting or conferences, such as Zoom, but with greater quality.

At present, the service operates in DEMO mode and allows one to process files downloaded from their computer with some restrictions. For example, the length of the video is restricted to 10 seconds. If the video is longer, then the service will cut it. File size should be smaller than 75 megabytes, only popular file formats can be uploaded (.mp4, .mkv, .mpg, .avi).

How it works

The heart of the technology is a neural convolutional network trained by our specialists. It calculates the alpha channel for each frame, where the human area is opaque, and all other areas are transparent. The final image quality depends on the accuracy of the training and network settings. For this reason, we spend a lot of effort improving this technology. More information about network training is available in our website as well as in several articles in our blog (see References).

Original image
Mask from model
Foreground
Final image

New features in the updated version

If you used the previous version of our service, then you will surely be interested in this section. Our specialists accelerated the previous model and the processing speed increased by about 1.5–2 times from 4–5 fps to 8–10 fps. At the same time, the output picture did not experience any quality decline. On the contrary, quality with regards to the processing of borders and translucent objects such as hair improved. We also reduced the amount of noise and dropouts of regions.

That’s all for now. Thank you for your attention. I will finish the article with several videos as examples of our service.

References

https://bgrem.deelvin.com/

https://medium.com/deelvin-machine-learning/human-image-segmentation-experience-from-deelvin-5148a6cc71da

https://medium.com/deelvin-machine-learning

--

--