BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Write once\, run anywhere inference with GStreamer
DTSTART;VALUE=DATE-TIME:20261010T092500Z
DTEND;VALUE=DATE-TIME:20261010T095500Z
DTSTAMP;VALUE=DATE-TIME:20261010T020840Z
UID:indico-contribution-603@indico.freedesktop.org
DESCRIPTION:Speakers: Sebastian Dröge (Centricular Ltd)\nModern devices c
 ome with multiple different hardware components that can be used for infer
 ence workflows: CPU\, NPU\, iGPU\, dGPU\, etc. The APIs to access them dif
 fer not only between the individual components and hardware vendors\, but 
 also between different platforms. To complicate the situation further\, mu
 ltiple hardware vendors provide extension libraries on top of GStreamer fo
 r accessing their specific hardware as efficiently as possible.\n\nWhen wr
 iting applications\, developers either have to implement a custom abstract
 ion layer or\, more often\, lock into a single hardware/platform which inc
 reases maintenance costs and limits deployment.\n\nSince the Analytics API
  was introduced to GStreamer in 1.24\, this situation has considerably imp
 roved. Many inference framework-agnostic tensor decoders for different mod
 els were introduced\, as well as new model-agnostic metadata for inference
  results and overlay and tracking elements. Together with the onnxinferenc
 e and tfliteinference elements and the newly added whisper.cpp and llama.c
 pp elements\, GStreamer can be used to develop write once\, run anywhere i
 nference pipelines.\n\nThis talk will provide an overview of the current s
 tate of inference in GStreamer\, discuss possible future developments\, an
 d show how GStreamer can be used to develop cross-platform inference pipel
 ines.\n\nhttps://indico.freedesktop.org/event/14/contributions/603/
LOCATION:Impact Hub Prague
URL:https://indico.freedesktop.org/event/14/contributions/603/
END:VEVENT
END:VCALENDAR