Apr 12 2019
Kandao, the leading 360/VR imaging & camera solution company, today introduced a new feature called 'AI Slow-motion 'that can turn 30 fps regular videos into max 300 fps super slow motions, using the machine learning system.
The new feature is firstly applied into Kandao Obsidian and QooCam, the industry-leading 360 cameras, making both upgraded into cameras that enable to create high-quality slow motion.
"Slow-motion offers possibilities to capture special moments in an epic way, but not all camera can do it. The high-fps camera is very expensive, as it requires large memories and is data-intensive, which is even more challenging for 360 cameras with multiple lenses." Said Dan Chen, Kandao CEO.
The sample video:https://youtu.be/E-1zfKCfQzU
Now, by applying the neural network to predict the extra intermediate frames, Kandao Obsidian and QooCam user can create a smooth, cinematic high-fps slow motion from the existing 360/VR video while maintaining the best quality.
Compared with the optical flow method, Kandao says the biggest advantage of using machining learning to slow down the footage is the frame interpolation can be more accurate, less jagged edges or artifacts, thanks to the neural network. It is the first time the latest slow-motion computer vision research applies to commercial use, with lighter system requirement and faster processing speed.
By using the new feature, users can slow down the footage to up to 10x slow motion after taking them. For example, a 360-degree video originally captured at 8k 30fps can be transferred into a 300fps slow-motion in 8K, a 4k 60fps video can be exported at 4K 600fps, by selecting a factor of ten during the 360 stitching workflow in Kandao/QooCam Studio software.
Together with the 'AI slow motion', Kandao also released a range of software upgrades in Kandao Studio V3.0 for the 8K 3D VR camera Obsidian, including the ultra-smooth video stabilization, stitching speed optimization and direct connection to export files from camera, etc.
Source: https://www.kandaovr.com/