-
project slides https://t.csdnimg.cn/hSSe
-
project video https://edu.csdn.net/huiyiCourse/detail/1202
- SparkleTime can record highlight moment with a streaming video. This usage is based on OpennVINO tech support. Details are WIP.
-
Current version supports for highlighting three games ( PUBG, LOL and World of Tank)
-
Before run the app, please go to this link and find source video. You can also check the sparkleTime video colletected from the video source If you cannot access the folder, please feel free to email yaru.du@intel.com
| Game name | source video | sparktletime video |
|---|---|---|
| PUBG | .\PUBG\test_PUBG.mp4 | .\PUBG\game_highlights_demo.mp4 |
| LOL | .\LOL\test_lol.mp4 | .\LOL\sparkletime_lol.mp4 |
| World of Tank | .\World_of_Tank\test_WoT_1.mp4 | .\World_of_Tank\sparkletime_wot.mp4 |
- python3
- openvino 2020.1 version. Doing steps till
run setupvars.batis enough- please refer to OpenVINO@intel windows installation
- please refer to OpenVINO@intel linux installation
Use anaconda to manage python environment is recommended, Here is some installation reference.
- To run World of Tank demo, please use the following command.
python game_highlight.py -i testinput/test_PUBG.mp4 -o output -g PUBG
- To run World of Tank demo, please use the following command.
python game_highlight.py -i testinput/test_lol.mp4 -o output -g LOL
- To run World of Tank demo, please use the following command.
python game_highlight.py -i testinput/test_WoT_1.mp4 -o output -g WoT
The sparkletime video will be saved in the output folder
| Model type | Model framework | Model size | OpenVINO supported |
|---|---|---|---|
| Text detection | Pixel_link + MobileNet v2 | 25.7MB | FP32/FP16/INT8 @CPU, FP32/FP16@GPU |
| Text recognition | LSTM+CTC Loss | 45.3MB | FP32/FP16@CPU, FP32/FP16@GPU |
- test on one instance
| \ms | AMD3700U CPU | ICELAKE CPU |
|---|---|---|
| text detection tf | 3161.35 | 1541.52 |
| text recognition tf | 5286.16 | 3002.85 |
| muscidance tf | 594.9 | 238.61 |
| musicdance OV FP32 | 216.03 | 68.54 |
| musicdance OV INT8-FP32 | 144.8 | 58.3 |
The performance of FP32 model on intel cpu is better than on AMD cpu benefits from the hardware, not from OpenVINO. Becuase if tensorflow model is tested on both HW, it also has the same gap.