isabella232 / dlafs

---

Deep Learning Accelerator Framework for Stream (DLAFS) project

Deep Learning Accelerator Framework for Stream (DLAFS) is based on Gstreamer. Leveraging MediaSDK, OpenCL, OpenCV and OpenVINO, DLAFS is able to perform video codec, computer vision processing, deep learning processing (including but not limited to object detection and classification/recognition).

0. Compile and install

---

Refer to DLAFS_Release_Notes_2020R3.pdf for detailed steps.

1. Run a test case

---

cd scripts/1.16.2
./prepare_omz_resource.sh
source setup_test
cd ../test/1.16.2
./uc4.sh
Several JPEG file as well as metadata.json should be created.

2. Best known configuration

---

CPU: Intel(R) Core (TM) i7-1185GRE CPU
OS:  Ubuntu 18.04.4
Kernel: lts-v5.4.49-yocto
OpenVINO: 2020.4.287
OpenCL SDK: 2019.5.345
Gstreamer: OpenVINO 2019.4.287
OpenCV: 4.4.0-openvino with Intel VA support
JSON-C: f8c632f579c71012f9aca81543b880a579f634fc
Libva: 2.8.0
Intel GMMLib: 20.2.2
Intel Media Driver: 20.1.1
Intel Media SDK: 20.2.0
OpenCL Runtime: 20.25.17111

CPU: Intel(R) Core (TM) i7-8700K/i7-7700K/i7-6700K CPU
OS:  Ubuntu 18.04.4
Kernel: 5.3.0
OpenVINO: 2020.4.287
OpenCL SDK: 2019.5.345
Gstreamer: OpenVINO 2019.4.287
OpenCV: 4.4.0-openvino with Intel VA support
JSON-C: f8c632f579c71012f9aca81543b880a579f634fc
Libva: 2.8.0
Intel GMMLib: 20.2.2
Intel Media Driver: 20.2.0
Intel Media SDK: 20.2.0
OpenCL Runtime: 20.25.17111