GStreamer AI commands don't work with Debian 13 - qtimlvconverter error

Issues
1

I see the gst-ai-* commands installed under the Debian 13 image and tried them but they fail with errors.

$ gst-ai-object-detection -s bus.jpg -t 3 --model=v2.32_QCS6490/yolonas.dlc --labels=labels/yolonas.labels

Fails with error;

TARGET Can support file source, RTSP source and camera source
File Source is Selected
Setting DSP as default Runtime
Running app with model: v2.32_QCS6490/yolonas.dlc and labels: labels/yolonas.labels
Failed to create qtimlvconverter
ERROR: failed to create GST pipe.

dmesg shows related errors of;

[ 9772.217193] [    T747] msm_vidc:     fw: <VFW_E:HostDr:unkn:--------:--> VenusHostDriver_ParseC2Command(928): QC_IMAGE_VERSION_STRING=video-firmware.2.4.2-dcbb0b7fae0340044243099914ccd743e3259faa
[ 9772.217210] [    T747] msm_vidc:     fw: <VFW_E:HostDr:unkn:--------:--> VenusHostDriver_ParseC2Command(929): IMAGE_VARIANT_STRING=PROD
[ 9772.217218] [    T747] msm_vidc:     fw: <VFW_E:HostDr:unkn:--------:--> VenusHostDriver_ParseC2Command(930): OEM_IMAGE_VERSION_STRING=hyd-lablnx446
[ 9772.217227] [    T747] msm_vidc:     fw: <VFW_E:HostDr:unkn:--------:--> VenusHostDriver_ParseC2Command(931): BUILD_TIME: May 20 2024 18:00:28
[ 9772.222029] [    T747] msm_vidc:     fw: <VFW_E:HostDr:unkn:--------:--> VenusHostDriver_ParseC2Command(928): QC_IMAGE_VERSION_STRING=video-firmware.2.4.2-dcbb0b7fae0340044243099914ccd743e3259faa
[ 9772.222050] [    T747] msm_vidc:     fw: <VFW_E:HostDr:unkn:--------:--> VenusHostDriver_ParseC2Command(929): IMAGE_VARIANT_STRING=PROD
[ 9772.222063] [    T747] msm_vidc:     fw: <VFW_E:HostDr:unkn:--------:--> VenusHostDriver_ParseC2Command(930): OEM_IMAGE_VERSION_STRING=hyd-lablnx446
[ 9772.222076] [    T747] msm_vidc:     fw: <VFW_E:HostDr:unkn:--------:--> VenusHostDriver_ParseC2Command(931): BUILD_TIME: May 20 2024 18:00:28
[ 9772.222419] [   T1951] CAM_ERR: CAM-CPAS: cam_cpas_subdev_ioctl: 750: Invalid command -2140645888 for CPAS! from pid 1951, comm gst-ai-object-d 
[ 9772.222452] [   T1951] CAM_ERR: CAM-CORE: cam_subdev_ioctl: 84: Invalid command -2140645888 for cam-isp from pid 1951 comm gst-ai-object-d 
[ 9772.222471] [   T1951] CAM_ERR: CAM-CORE: cam_subdev_ioctl: 84: Invalid command -2140645888 for cam-jpeg from pid 1951 comm gst-ai-object-d 
[ 9772.222486] [   T1951] CAM_ERR: CAM-CORE: cam_subdev_ioctl: 84: Invalid command -2140645888 for cam-lrme from pid 1951 comm gst-ai-object-d 
[ 9772.222547] [   T1951] CAM_ERR: CAM-SENSOR: cam_sensor_subdev_ioctl: 68: Invalid ioctl cmd: -2140645888 
[ 9772.222558] [   T1951] CAM_ERR: CAM-SENSOR: cam_sensor_subdev_ioctl: 68: Invalid ioctl cmd: -2140645888 
[ 9772.222951] [   T1951] CAM_ERR: CAM-MEM: cam_mem_mgr_request_mem: 1544: failed. mem_mgr not initialized 
[ 9772.222953] [   T1951] CAM_ERR: CAM-ICP: cam_icp_allocate_hfi_mem: 2887: Unable to allocate qtbl memory 
[ 9772.223342] [   T1951] CAM_ERR: CAM-ICP: cam_icp_subdev_open: 97: FW download failed 
[ 9772.223372] [   T1951] CAM_INFO: CAM-SMMU: cam_smmu_need_force_alloc_cached: 383: force_cache_allocs=1 
[ 9772.223377] [   T1951] CAM_ERR: CAM-MEM: cam_mem_mgr_get_dma_heaps: 480: qcom,display heap not found, rc=0 
[ 9772.223396] [   T1951] CAM_WARN: CAM-CRM: cam_req_mgr_close: 189: release invoked associated userspace process has died, open_cnt: 1 
[ 9772.223400] [   T1951] CAM_ERR: CAM-CORE: cam_subdev_ioctl: 79: shutdown device failed(rc = -22) for cam-isp 
[ 9772.223402] [   T1951] CAM_ERR: CAM-CORE: cam_subdev_ioctl: 79: shutdown device failed(rc = -22) for cam-lrme 
[ 9772.223403] [   T1951] CAM_ERR: CAM-CORE: cam_subdev_ioctl: 79: shutdown device failed(rc = -22) for cam-jpeg

There is a post about camera not working on Debian 13, however as the gst-ai- command is being run to process a file, I don’t see why its attempting to access camera?

When is new Debian image due to be released with a fix?

2

however as the gst-ai- command is being run to process a file, I don’t see why its attempting to access camera?
→
The gst-ai- commands are typically AI samples related to video processing, which require a video input source.
The options for video input sources include:

  1. Real-time image data from a MIPI camera (this is also the default scenario)
  2. Pre-saved MP4 files,
  3. RTSP data streams.
3

The same error occurs if feeding it an MP4 video file.

$  gst-ai-object-detection -s palace.mp4 -t 3 --model=v2.32_QCS6490/yolonas.dlc --labels=labels/yolonas.labels

TARGET Can support file source, RTSP source and camera source
File Source is Selected
Setting DSP as default Runtime
Running app with model: v2.32_QCS6490/yolonas.dlc and labels: labels/yolonas.labels
Failed to create qtimlvconverter
ERROR: failed to create GST pipe.
4

We expect the new Debian image to be released in June. Please stay tuned for our announcements for updates .

5

This issue is already known to us, and the current Debian version does not support this feature, so you do not need to continue attempting it.
Of course, you are also welcome to develop it on your own.

6

Ok. In the Debian image are the drivers for the NPU installed? So can we use the Qualcomm AI Engine Direct SDK with it?

7

In the Debian image are the drivers for the NPU installed?
–>Yes, the Debian image has the NPU drivers installed.
So can we use the Qualcomm AI Engine Direct SDK with it?
–>Yes, you can try it.

8

I think it would be better to use Qualcomm Linux than Debian Linux, because the manual, demos, etc. of Rubik Board are all explained based on Qualcomm Linux. I am also working on Qualcomm Linux.

9

Dear customer,
Welcome to Qualcomm Linux.
Additionally, we will release the Debian user-manual in June, maybe earlier.