Asset Integration
As the Vionlabs Discovery Platform is primarily focused on the processing of media AV content in order to generate rich data, it is essential for most use cases that we can process raw media files for our partner catalogs. This section provides information on the options for integration of asset processing.
In all cases it is a requirement that a URI for the file be provisioned in the catalog information to enable the Vionlabs Discovery Platform to locate media assets and associate them with the correct catalog item.
Media File Requirements
The Vionlabs Discovery Platform provides maximal flexibility in terms of formats and containers for video files and all major formats are supported. This includes, but is not limited to: - MP4 (.mp4) - Matroska (.mkv)
In terms of audio, we do not perform any language specific processing and so there are no language requirements applied.
It is important that all assets are provided with audio and video included in a single file.
For video, Vionlabs requires medium to high resolution content. Resolutions between 720p and 4K are sufficient for our AI algorithm to capture the most important features and insights from the content (we do not want mezzanine-sized files as the additional size and information does not add anything to the analysis). All standard frame rates and resolutions are supported with automatic detection and normalization.
Cloud-Based Processing
Due to the computationally intensive nature of processing AV files with advanced AI algorithms, Vionlabs' preferred approach is to process media assets in the cloud. This allows the platform to take advantage of horizontal scalability and flexible architecture availability to maximise the efficiency of the required operations and workflows.
For cloud-based processing, transfer of partner video assets to the Vionlabs Discovery Platform can be integrated in one of two ways:
- Partner push integration
- Vionlabs pull integration
With the push integration, media files are transferred to a Vionlabs-provisioned Google Cloud Storage bucket by the partner. The upload bucket details and credentials for a Google Service Account with appropriate permissions will be provided by your Vionlabs integration lead. In order to ensure security of your data, minimal permissions are provisioned to the service account. It is only possible to upload files to the bucket. All other bucket operations are disabled.
With the pull integration, the Vionlabs Discovery Platform will automatically download the media asset file from a URL. The URL must be provisioned in the asset information fields of the catalog in order to facilitate this.
On-Premises Processing
On-premises processing is facilitated by dedicated subsystems within the Vionlabs Discovery Platform. In this scenario the first layer of processing is carried out on CPU resources on the customer premises. The results of this stage of processing are then transferred to the cloud for further stages of processing.
The Vionlabs on-premises processing system is designed with the assumption that it is a requirement to be able to operate in a fashion that is "air-gapped" from the rest of the partner's processing infrastructure. The top-level requirements for on-premises processing are:
- Provision of an adequate level of compute resources relative to the size of the catalog (please consult your Vionlabs integration lead for discussions on appropriate specifications). Compute resources can be VMs or bare metal machines. GPUs are advantageous for this layer of processing.
- A Linux operating system installed on all compute instances (Ubuntu, CentOS or RHEL preferred)
- A VPN endpoint that can be used to programmatically connect to the compute instances (preferably OpenVPN). Appropriate credentials must be provided. Your Vionlabs integration lead can provide details of a fixed IP address, should you wish to employ IP whitelisting.
- A small number of ports open on the instance for access via the VPN.
- Access to object storage or NFS mounts for the required files.