Most images today are processed with AI/ML, where differences in pixel values are critical and may be invisible to the human eye. ML model reliability depends on high-quality data, best achieved with raw images.
Raw data’s size is a bottleneck for transmission speed, processing, storage and associated costs.
Image compression could tackle these, but, currently impose limitations:
More details, (higher dynamic range), and overall maximum information facilitates enhanced ML model training, which can improve generalization of the models.
While undetectable to the human eye, compression artifacts introduced by conventional methods can adversely affect the performance of ML models.
Apply pre-processing techniques, e.g. noise reduction, normalization, etc. without introducing bias or artifacts to experiment with diverse pre-processing criteria tailored to your AI/ML application
Jetraw Core is in-camera raw compression combining the size of lossy formats with the reliability and quality of lossless technologies.
It applies lossless compression to the signal component, following image preparation with lossy compression of the noise. This achieves the highest ratio and the highest quality output on the market.
Jetraw turns data into measurement by embedding a custom noise model
based on the sensor’s physical properties. This enables image quality validation and optimization for AI/ML.
High-volume high-quality data required for ADAS and AD system’s reliability gets in the way of data throughput, also demanding high resource use within vehicular environments. This results in substantial costs. Conventional lossless compression techniques struggle with scalability due to limited speed and efficiency.
FPGA or ASIC-embedded Jetraw Core allows for low latency cloud-based RAW video processing for robust ADAS and AD systems development. With an option to be also implemented as a software, it enhances data transmission speed and reduces computing needs. Jetraw Core lowers energy consumption, optimizes data storage, and leverages existing IT infrastructure for greater throughput at reduced bandwidth. Furthermore, it enhances image reliability for AI and ML processing.
For a system with 1 x 8Mpx cameras, 36fs, with Jetraw Core compression, Tier 1 client enables 6x higher data throughput for cloud compared to original raw, reduces transmission time, and saves 83% on costs.
The aerospace industry grapples with slow image acquisition and large data volume from on-board sensors and imaging systems, preventing high throughput for space-to-ground data transmission. Efficient data handling is vital in enhancing satellite imaging and telemetry.
Jetraw Core integrated in FPGA IP Core increases data transmission speed, preserving highest resolution. This enables raw imaging for scalable and reliable AI analysis.
It reduces latency, energy consumption, and maximizes existing infrastructure without the need to upgrade, resulting in a better throughput even at reduced bandwidths. FPGA- or ASIC- implemented compression can be updated remotely, adapting to the needs of reconfigurable functionalities and end users’ needs of high-quality images.
Jetraw Core compression in FPGA helps companies like European Space Agency increase throughput of images and volume of data to store on the onboard RAM 6x, facilitate processing and increase download speed 6x compared to original raw. On the ground, it helps to streamline image data distribution and reduce costs by over 80%.
Camera manufacturers and OEMs often face a trade-off between resolution, speed of acquisition, and bandwidth. High-resolution imaging demands more bandwidth and compromises the acquisition speed, restricting the performance of imaging systems. This constraint limits the effective use of camera hardware, reducing the quality and the amount of data acquired, preventing the use of camera at its full potential.
There are several ways camera manufacturers can benefit from Jetraw Core depending on where it’s integrated and whether it’s embedded on FPGA or as a software library. Jetraw Core can facilitate:
1. Higher frames per second.
2. Higher resolution, maintaining the same hardware
3. Higher bit depth, keeping existing speed, and
4. Faster download speed, for a given resolution. Decompression is adapted to the needs of end-user’s image acquisition pipeline.
Excelitas PCO enable higher data transmission speed for their camera users, whilst preserving highest resolution. This allowed them to avoid compromises, offer camera at a competitive price, maintain existing hardware while enhancing performance for the end-users.