As announced in the previous blog post, the IVA product (IBM Intelligent Video Analytics) was withdrawn from the market.
Its successor is IBM Video Analytics. Yes, this is another IVA, and so far the trend seems to be for people to call it VA or IVA_I tried calling it SIVA (Son of IVA) for a while but that never caught on.
So what's new, what's different, and what has been retained?
- The first thing you'll notice in the new offering (I'm going to call it VA 1.0) is that the UI has been given a face-lift. It has a new color-scheme but is for the most part the same UI as in IVA 3.0_ same layouts, navigation, etc so those familiar with IVA 3.0 will find it easy to navigate.
- As far as real new functionality, there are several significant improvements including:
- DLE Alert Filtering
- This feature uses Deep Learning models running on the DLE to verify alerts generated by traditional behavioral alerting. In initial testing we have found this can greatly reduce the number of false positive alerts while not putting a large load on the (expensive) GPUs that the DLE uses. This combination of traditional video analytics with Deep Learning is one advantage that retention of previous IVA technology provides. In short we can deliver more cost effective solutions while still leveraging the exceptional improvement in accuracy that Deep Learning provides.
- Camera Import
- This feature (supported for Milestone and Genetec VMS's today) greatly reduces the administrative workload when populating large populations of cameras in VA as compared to IVA.
- TensorRT Optimization
- This feature allows Deep Learning models (including those trained with IBM Power AI Vision) to be optimized so they run more efficiently on the GPUs. We have seen a significant increase in the frame rate of models with insignificant loss of accuracy using TensorRT, so this is another thing that will reduce the hardware costs of deployments.
- There is a new RESTful API called the SSE Gateway API that provides access to the administrative and operational functions of the SSE using REST calls
- There is a new RESTful API to the Camera Import functions
- Swagger docs
- Both the SSE Gateway and Camera Import APIs have Swagger docs
- Swagger (and OpenAPI) will be how we deliver all API documentation in the foreseeable future.
- DLE Alert Filtering
- No Facial Recognition in OOB product. This change was driven by the corporate responsibility initiative that led to the end of IVA.
- Facial Recognition will still be available as an add-on module which will be provided only to customers and markets that pass review to ensure that this technology is used in an ethical and socially responsible manner.
- No Ethnicity Classifications
- We're not sure whether this functionality will be made available as an add-on module or not.
- As mentioned above, the UI is very similar to the one in IVA
- The Sub-Systems (MILS, SSE, DLE) are the same but MILS and SSE now stand for different names_ Metadata Ingestion Lookup and Signaling (MILS) and Semantic Streams Engine (SSE)
- All Traditional Analytics are retained with the exception of skin tone classification. These are still valuable both stand-alone and in combination with Deep Learning
- The team. Most of the people working on the new VA product are the same ones who worked on IVA. So product knowledge, relationships with customers, etc have a lot of continuity.