amadeus-zte
amadeus-zte
Adlik should support custom-defined ops in openvino, such as "tensorflow_use_custom_operations_config".
As far, Adlik serving need to use Adlik compiler to convert model format and create config file which is needed when running inference. However, if users have already runtime model...
One case is ansor.
Demo board is kria260 with xczu5ev chip, in which one DPU (B4096) is used. Scenario is cloud deployment.
Try to run faster-rcnn in adlik with openvino and tensorRT. Original model reference: https://github.com/facebookresearch/detectron2 R50-FPN
Please provide formal external interface document.