- Docker image:
nvcr.io/nvidia/deepstream:5.0-dp-20.04-samples
- API: gst-python + pyds
- Hardware: nvidia-smi output =
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.82 Driver Version: 440.82 CUDA Version: 10.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 1660 Off | 00000000:1E:00.0 Off | N/A |
| 29% 32C P8 9W / 120W | 160MiB / 5944MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
+-----------------------------------------------------------------------------+
I have a pipeline like this:
srcs -> nvstreammux -> nvinfer -> nvtracker -> nvvideoconvert(1) -> capsfilter(1) -> nvmultistreamtiler -> nvvideoconvert(2) -> nvdsosd -> tee -> sinks
- When capsfilter(1) is set to
"video/x-raw(memory:NVMM),format=RGBA"
- setting tiler
show-source
to any value != -1 makes the display black instead of showing the respective source. - setting tiler
show-source
back to -1 then all sources are tiled and displayed normally.
- setting tiler
- When capsfilter(1) is left empty, then tiler
show-source
works as expected but the problem is that in this case images can’t be extracted from buffer probe on nvvideoconvert(1), becausepyds.get_nvds_buf_surface
works only with RGBA.