You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Sep 30, 2024. It is now read-only.
Dependency: MediaSDK 20.3 and OpenVINO 2021.1
New features on R3:
Support CPU inference by option “-infer::device GPU”.
Support shared inference network instance between sessions
Support setting decoding output to RGBP by option “-dc::rgbp”
Support 2 video decoder outputs from SFC and VPP in different sizes and color format
Support more detection networks by specify XML files
Signed-off-by: Elaine Wang <[email protected]>
* Intel® platforms supported by the MediaSDK 20.1.1 and OpenVINO 2020.3.
48
+
* Intel® platforms supported by the MediaSDK 20.3.0 and OpenVINO 2021.1.
49
49
* For Media SDK, the major platform dependency comes from the back-end media driver. https://github.com/intel/media-driver
50
50
* For OpenVINO™, see details from here: https://software.intel.com/en-us/openvino-toolkit/documentation/system-requirements
51
51
52
52
# How to build
53
53
54
54
Run build_and_install.sh to install dependent software packages and build sample application video_e2e_sample.
55
55
56
-
Please refer to ”Installation Guide“ in [user guide](./doc/concurrent_video_analytic_sample_application_user_guide_2020.2.0.pdf) for details.
56
+
Please refer to ”Installation Guide“ in [user guide](./doc/concurrent_video_analytic_sample_application_user_guide_2020.3.0.pdf) for details.
57
57
58
58
## Build steps
59
59
@@ -68,10 +68,10 @@ cd cva_sample
68
68
```
69
69
This script will install the dependent software packages by running command "apt install". So it will ask for sudo password. Then it will download libva, libva-util, media-driver and MediaSDK source code and install these libraries. It might take 10 to 20 minutes depending on the network bandwidth.
70
70
71
-
After the script finishing, the sample application video_e2e_sample can be found under ./bin. Please refer to "Run sample application" in [user guide](./doc/concurrent_video_analytic_sample_application_user_guide_2020.2.0.pdf) for details.
71
+
After the script finishing, the sample application video_e2e_sample can be found under ./bin. Please refer to "Run sample application" in [user guide](./doc/concurrent_video_analytic_sample_application_user_guide_2020.3.0.pdf) for details.
72
72
73
73
# Known limitations
74
74
75
-
The sample application has been validated on Intel® platforms Skylake(i7-6770HQ), Coffee Lake(i7-8559U i7-8700) and Whiskey Lake(i7-8665UE).
75
+
The sample application has been validated on Intel® platforms Skylake(i7-6770HQ), Coffee Lake(i7-8559U i7-8700), Whiskey Lake(i7-8665UE) and Tiger Lake U(i7-1185G7E, i5-1135G7E).
## Can I use other OpenVINO version rather than 2020.3 ?
48
48
Yes, but you have to modify some code due to interfaces changing. And also you need to download the IR files and copy them to ./model manually. Please refer to script/download_and_copy_models.sh for how to download the IR files.
49
+
50
+
## When run 4 channel decode plus inference and display on APL, the CPU occupy ratio is very high and fps is low
51
+
Please refer to par file par_file/inference/n4_face_detection_rgbp.par. It uses option "-dc::rgbp" that make the SFC outputs RGB4 for display and VPP outputs RGBP for inference input. Then there is no need to use OpenCV for resizing and color conversion which consume much CPU time on APL.
52
+
Note, "-dc::rgbp" only works with "-infer::fd". Will support more inference types in the future.
0 commit comments