Slow FPS in Deepstream Python test applications over Jetson Nano

Hi everyone,

This is my device information:
Hardware Platform:Jetson nano 4GB
deepstream-app version 5.1.0
DeepStreamSDK 5.1.0
CUDA Driver Version: 10.2
CUDA Runtime Version: 10.2
TensorRT Version: 7.1
cuDNN Version: 8.0

I’m trying to run deepstream-test-2 application over Jetson Nano 4GB but getting FPS of only 6.8 avg. I had updated the config file according to Nano and also had created the engine files. I had tested the engine files with trtexe and it is fine. I had checked the inference time manually and found that it is around 2.1457672119140625e-05 per frame that means the inference is working fine.
I had tried this but it didn’t help.

I had checked the debugger as well. Provided below are the warnings that I have got so far:

0:00:00.527116941 12786 0x31676c40 WARN omx gstomx.c:2826:plugin_init: Failed to load configuration file: Valid key file could not be found in search dirs (searched in: /home/apoorv/.config:/etc/xdg/xdg-unity:/etc/xdg as per GST_OMX_CONFIG_DIR environment variable, the xdg user config directory (or XDG_CONFIG_HOME) and the system config directory (or XDG_CONFIG_DIRS)
0:00:00.968608426 12786 0x31676c40 WARN v4l2 gstv4l2object.c:4434:gst_v4l2_object_probe_caps:nvv4l2-decoder:src Failed to probe pixel aspect ratio with VIDIOC_CROPCAP: Unknown error -1
0:00:00.968701032 12786 0x31676c40 WARN v4l2 gstv4l2object.c:2375:gst_v4l2_object_add_interlace_mode:0x314145f0 Failed to determine interlace mode
0:00:00.968770253 12786 0x31676c40 WARN v4l2 gstv4l2object.c:2375:gst_v4l2_object_add_interlace_mode:0x314145f0 Failed to determine interlace mode
0:00:08.665443038 12786 0x31676c40 WARN basesrc gstbasesrc.c:3583:gst_base_src_start_complete: pad not activated yet
NvMMLiteOpen : Block : BlockType = 261
NVMEDIA: Reading vendor.tegra.display-size : status: 6
NvMMLiteBlockCreate : Block : BlockType = 261
0:00:08.904337434 12786 0x31756b70 WARN v4l2 gstv4l2object.c:4434:gst_v4l2_object_probe_caps:nvv4l2-decoder:src Failed to probe pixel aspect ratio with VIDIOC_CROPCAP: Unknown error -1
0:00:08.904485198 12786 0x31756b70 WARN v4l2 gstv4l2object.c:2375:gst_v4l2_object_add_interlace_mode:0x314145f0 Failed to determine interlace mode
0:00:08.904629837 12786 0x31756b70 WARN v4l2 gstv4l2object.c:2375:gst_v4l2_object_add_interlace_mode:0x314145f0 Failed to determine interlace mode
0:00:08.923893307 12786 0x31756b70 WARN v4l2videodec gstv4l2videodec.c:1673:gst_v4l2_video_dec_decide_allocation: Duration invalid, not setting latency
0:00:08.927560799 12786 0x31756b70 WARN v4l2bufferpool gstv4l2bufferpool.c:1065:gst_v4l2_buffer_pool_start:nvv4l2-decoder:pool:src Uncertain or not enough buffers, enabling copy threshold
0:00:08.939178341 12786 0x563138a0 WARN v4l2bufferpool gstv4l2bufferpool.c:1512:gst_v4l2_buffer_pool_dqbuf:nvv4l2-decoder:pool:src Driver should never set v4l2_buffer.field to ANY

I had also tested the C-based test applications as well and facing the same issue of low FPS.

Please can anyone help me out with this issue?

Hi,
Sorry for the late.
Can you specify how you measure FPS for test2 sample? and the issue is from builtin model or your model?

Hi @amycao ,
I have used Deepstream-python-demos repositories GETFPS function to measure the FPS. I have not used custom model, its resnet10 that I’m using here. If you need anymore details please let me know.

I see you observe low fps with c version too, so i try to repro your issue with c version test2 sample, with fps.patch applied, i can get real time performance. please let me know more info.

Frame Number = 1440 Number of objects = 8 Vehicle Count = 6 Person Count = 2
Frame Number = 1441 Number of objects = 0 Vehicle Count = 0 Person Count = 0
End of stream
Returned, stopping playback
Average fps 133.911247
Deleting pipeline

fps.patch (1.1 KB)

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.