Video Intelligence Cloud Service
The most efficient AI video analytics
Bring serenity to your connected world. So you can focus on what matters the most.
Optimize profit and consume less energy. Bring more cost-effective services to the mass market.
Even create new products that are not technically feasible before.
World's firsts
Directly analyze the entire video as a whole
Recognize each object only once
No need to detect each object frame by frame
Directly analyze encoded data
of compressed video
without decoding
Extract representative images
of all interesting objects
quickly and accurately
300X more efficient than GPU
- Such performance is equivalent to 53 Nvidia GeForce GTX TITAN X GPU.
- Google reports that the running time of mask_rcnn_inception_v2_coco is 79 ms per 600×600 image using an Nvidia GeForce GTX TITAN X GPU. To estimate the running time per 4M-pixel (2688×1520) image, the original resolution can be divided into 600×600 subsections overlapping by 300 in both vertical and horizontal dimensions, total 32 subsections, which guarantees that any object whose size is smaller than overlapping dimensions of 300×300 will be contained within at least one subsection. Yet an object larger than 300×300 may be not contained within any subsection and may be missed. To deal with larger objects, a down-sampled image at ¼ resolution is used. In total, 33 runs of mask_rcnn_inception_v2_coco are needed to process a 4M-pixel image, totaling 2.6 seconds. To process above 4M-pixel real-time video at full 20 fps, it takes 53 Nvidia GeForce GTX TITAN X GPUs to process in parallel.
- Using today's most cost-effective GPUs would require 11 NVIDIA GeForce RTX 3090 GPUs paired with 3 AMD Threadripper 3960X CPUs.
- As of February 2022, the most cost-effective GPU for AI is the NVIDIA GeForce RTX 3090 whose single-precision throughput is 29280 GFLOPS and release price is $1500. Since it takes 2.6 seconds to analyze a 4-megapixel image using an NVIDIA GeForce GTX TITAN X whose single-precision throughput is 6144 GFLOPS, it is estimated that it takes 0.55 seconds using an NVIDIA GeForce RTX 3090. Then, to process real-time video at 20 frames per second, 11 NVIDIA GeForce RTX 3090 GPUs are required to operate in parallel.
- In a typical system, 11 NVIDIA GeForce RTX 3090 GPUs are paired with 3 AMD Threadripper 3960X CPUs whose release price is $1400.
- Total cost of 11 NVIDIA GeForce RTX 3090 GPU and 3 AMD Threadripper 3960X CPU is $20700.
300X more efficient than GPU
to analyze video in real time
Key features
- Typical neural network can only analyze images of up to 800 pixels x 800 pixels.
- Our neural network can directly analyze videos of any resolution. Whether it is HD, 4K, or even 8K.
- Typical home security camera with AI capability can only detect a person up to 10 meters away.
- Utilizing our video analytics cloud service, an ordinary camera can identify a person more than 100 meters away.
- Typical video analysis cloud service only analyzes the first frame of each second of video. However, each second of video usually contains 20 to 30 frames of images, so it is likely to miss critical situations.
- Our video analytics cloud service analyzes every frame of every second of video so nothing is missed.
Unlimited resolution
-
Typical neural network can only analyze
video downsampled to lower resolution
Typical neural network analyzes images only up to 800 x 800 pixels.
Videos with higher resolutions must be downsampled to low-resolution images before being analyzed using such neural network. The loss of details, blurring features, diminishing textures significantly hinder the accuracy and effectiveness of AI.
-
Videolytics neural network can analyze
videos of any higher resolutions directly
Our neural network can directly analyze videos of any resolution. Whether it is HD, 4K, or even 8K, there is no need to downsample. By analyzing each frame at its native resolution, we guarantee unparalleled accuracy in object detection, enabling you to see the full picture and make decisions with confidence.
Recognize object 10X farther
-
Typical cloud service cannot
recognize people 10 meters away
Typical video intelligence cloud service can analyze only downsampled low-resolution videos. It often fails to identify people far away, because the images lack the necessary detail, become pixelated, blur important features, reduce the scale of objects, lose context, and are prone to noise and distortion.
-
Videolytics cloud service empowers an ordinary camera to
identify people over 100 meters away
Our video intelligence cloud service can directly analyze high-resolution videos without any downsampling. Our state-of-the-art technology ensures that even people as far as 100 meters away are recognized with clarity and precision by just an ordinary camera. With our high-resolution solutions, distance is no longer a barrier.
Never miss a single frame
Typical cloud service
Typical video intelligence cloud service only analyzes the first frame of each second, likely to miss critical situation.
Videolytics cloud service
Our video intelligence cloud service analyzes every frame of every second, never missing anything.