Cloud-based deployment is the most popular choice of deployment for Computer Vision. It also has the lowest cost of entry, so it suits start ups, smaller businesses or simply where the business does not want to invest in their own on-premise hardware and software systems, which then require additional monitoring, maintenance and support. Cloud deployment is also the most flexible, as agreements and contracts can take into account fluctuations in processing volume demands, so there are no inherent costs during low volume periods because you are not running and maintaining your own server infrastructure, and cloud providers can easily increase resources to meet high-demand periods.
Cloud deployment simply involves the use of an API to send the media to be processed to the cloud engine. Once processed the data is extracted it is saved into a JSON or XML format and sent back via the same API, or accessible via the cloud system’s dashboard.
With On-premise deployment, all data is processed and stored on private servers located within your business premises. While this adds layers of cost and other support and maintenance overheads, there are certain industries in which privacy and self-containment are critical. Additionally, where businesses need to consistently process high volumes of media each day/month, running their own hardware infrastructure can be more cost-effective.
Another benefit of on-premise is the mitigation of latency because the visual media (which is inherently large) never leaves their own network and so is not affected by slower and unguaranteed public network bandwidths.
Implementation of on-device computer vision deployment enables any number of diverse applications where processing must occur directly on the device in situations where data connections to main servers are not guaranteed, or perhaps even wanted.
On-Device deployment can be implemented on any handheld device with a camera and processor, so any modern smartphone can become a computer vision capture and processing device.
This is a fantastic, flexible deployment option of Visual-AI technology that is best implemented in logistics and customs, or consumer engagement applications.
Where the device also incorporates internet connectivity, it can automatically upload processed data and download model updates when connected to the network, but still capture and process images when not connected or out of range.
Embedded deployment is ideal where there is a requirement for a very specific computer vision task to be implemented within a piece of hardware. This may be for a commercial product, such as network firewalls with built-in phishing detection, or a piece of hardware used in a manufacturing/production line to monitor quality issues or the status of robotic arms. However, in practice, the applications for embedded Computer Vision are limitless.
The right deployment option for your project depends entirely on how you intend to use computer vision technology within your organization and what your expectations are as well as some key considerations such as budget, space and security requirements. Here are some questions to ask when you are discussing deployment options with your team.
The higher the volumes of data to process, the more likely that an on-premise solution will be more cost-effective. But there is a cross-over point to consider as the cost of implementing and maintaining your own hardware infrastructure and storage is high, so you need to balance the costs of the cloud system (that has all infrastructure costs built into their processing fees) against the savings on processing fees if you host the infrastructure yourself.
Another factor is the consistency or your processing needs. Investing in hardware, whether physical servers or virtual servers in a private cloud, incurs a constant overhead, both for the hardware itself and technicians to manage it. So, the last thing you want is to have it sitting idle.
If you anticipate periods of low processing demand, you may be better off with cloud processing, where you can negotiate your sporadic usage into the contract.
Similarly, if you anticipate bursts of high processing demand for short periods, will you be able to quickly scale your hardware to keep up? In this instance, cloud processing may again be your best option because the provider can scale their solution faster than you can.
Many organisations, such as government bodies,the financial industry, healthcare organisations and some cybersecurity businesses, are bound by policy to retain, manage and process data on a secure, private server, as opposed to using cloud services and processors. Businesses like this tend to use private servers/clouds to store documents and digital correspondence; it’s likely it will also be essential that they do not use a public cloud server to process data.
Most business applications are platform to platform and constantly connected to a data network. For instance, a brand monitoring company will always operate from the same connected location, feeding data to and from the platform for users to analyse. Cloud deployment, on-premise deployment, is what you need.
However, if you operate a team of remote users, working in areas with little or no data connectivity at times, cloud and on-premise deployment will not serve your purpose at all. On-device deployment is the only option for you, allowing processing on a hand-held device without the need for data connectivity.
One reason cloud-based computer vision servers are so popular is because having an on-premise server will require a specialised team that can manage the server itself and understand the specifics of running computer vision systems, such as updating models.For some companies with the budget and organisational structure to do this, on-premise may be the answer. However in smaller companies or businesses that outsource their IT management, it might not be ideal to deploy computer vision on site. Cloud deployment means that you don’t have to worry about the maintenance or management of any additional servers, so it could be the best solution for you.
Pricing structures vary from provider to provider, so be sure to discuss this in depth with the computer vision supplier you choose to work with.
If you do not specifically require on-device or on-premise deployment but are exploring your options, it’s important that you choose the method that works best for your budget.
Cloud, on-premise and on-device deployment provides excellent flexibility for the widest range of applications, but that means it’s not lean. If you are looking to add a specific computer vision task to a hardware product, or add it into your production line, or even involved in robotics, then you need the solution to be as lean as possible and require only the computing power and memory to achieve that task and nothing else.
In this case embedded Computer Vision will be your best option. In this scenario a specific chip is designed to run the Computer Vision task required, with that chip then added into the product/hardware in question.
There are many questions you could ask yourself when it comes to choosing the best deployment option for you and your organisation. If you’re unsure of what will work in your favour, the most important thing to do is plan ahead before you talk to service providers and let them guide you towards the most sensible and effective solution for you.
Despite often being considered a “fledgling technology,” computer vision is already proving itself valuable in multiple scenarios. The success of computer vision […]
Press Release VISUA Solves Key Corporate Visual-AI Challenges With ‘Deploy-Anywhere’ On-Premise Solution Removing a key barrier to entry for large enterprise companies […]
PRESS RELEASE VISUA On-Device Deployment Takes World-Leading Visual-AI To Places It Couldn’t Go Before VISUA’s new On-Device implementation enables key applications for […]
Seamlessly integrating our API is quick and easy, and if you have questions, there are real people here to help. So start today; complete the contact form and our team will get straight back to you.