new technology
Infrared true color conversion algorithm ROMA
ROMA transcends the triple constraints of spatial similarity to ensure all-round transformation and upgrading, and the video transformation is clearer, smoother and more stable;
The multi-scale region discriminator based on VIT makes the transformation real and reliable, and the details are clear, giving people a more realistic visual experience.
The cross-spatial correlation constraint of time makes the video conversion consistent in time and space, smooth and flicker-free.
Global and local cross spatial correlation constraints make training faster and more stable.
Related papers have been accepted by ACM MM2022.
Principle block diagram of ROMA algorithm
Thermal imaging &; visible light
Video smooth conversion
There are many difficulties and challenges in transforming the black-and-white infrared world at night into the world seen by human eyes during the day: unavailable matching data, different imaging principles, different imaging styles, different scenes and so on. ROMA algorithm helps us to build a translation bridge, so that we can cross these obstacles and realize the clear and smooth conversion from night thermal imaging to daytime visible light.
The conversion effect is better.
More authentic and reliable infrared true color, better temporal and spatial consistency, smoother video experience, no flicker and no jump.
Triple upgrade of efficiency/detection/fusion
Efficiency upgrade: 10 times+training speed, extremely fast industrial deployment.
ROMA algorithm can complete model training very quickly, which is 1 1 times faster than F/LSSIM algorithm, and helps infrared true color to quickly log on to various platforms and devices.
Detection upgrade: more accurate night detection, security protection
After the conversion, the target is clearer, the detection is more accurate, ADAS is empowered, and multiple guarantees are safer.
Fusion upgrade: clearer video fusion
Infrared true color helps fusion, strengthens night vision, makes the scene clearer, makes fusion without artifacts, and makes night vision smoother.
By far the largest
Double-light city data set open source
The latest InfraredCity data set, that is, unparalleled night infrared and daytime visible videos in cities, highways and surveillance scenes, contains nearly 600,000 images, which is 20 times larger than IRVI.
This is the largest open-source dual-light city data set in the field of day-night infrared true color conversion, and the scenes are real and rich. Iresearch Optoelectronic is willing to join hands with industry partners to promote the development and application of infrared thermal imaging technology and further promote the prosperity and development of the industry.