With AI being all the rage these days and each vendor working on getting their wares to market with AI acceleration, besides Intel now having out their Meteor Lake CPUs that feature their Intel AI Boost (NPU), Intel is rather positioned well with their prolific open-source software contributions. One of the AI-related software contributions over the past week has been improvements to the FFmpeg multimedia library’s “DNN detect” filter for object detection within videos.
Back in 2021, Intel engineers originally worked on this “dnn_detect” filter for providing object detection within video content. Over the past week there’s been several patches merged to FFmpeg Git for further enhancing this deep neural network detection support.
Support for YOLOV3 was added to complement the existing SSD and YOLOV1V2 DNN detection model types. YOLOV3 was added since it can perform better on both large and small objects. YOLOV4 was also added.
Besides the new models there’s also been several other recent fixes/improvements to this dnn_detect video filter, thanks to Intel engineers.
With OpenVINO integration, the work can be accelerated not only on CPUs but also Intel’s GPUs and shiny new NPU. Thanks to Intel’s resources and numerous open-source talent going back years, they are primed for helping to enhance the open-source ecosystem to take advantage of their latest hardware capabilities.