Classification example, inference isn't working with custom model

[trr-reds]

Hello,
We have created a simple model respecting the inputs and outputs of your classification model. The program runs correctly after the flash but no inference results are received, the application seems to be stuck on the call to "pi_cluster_send_task_to_cl" in the "cam_handler" function.

Here is the nngraph at compile time

Here is the cfclient console

CPX: GAP8: *** Classification ***
CPX: GAP8: Starting to open camera
CPX: GAP8: Opened Camera
CPX: GAP8: Allocated memory for camera buffer
CPX: GAP8: Allocated memory for output
CPX: GAP8: Allocated memory for task
CPX: GAP8: Constructed CNN
CPX: GAP8: ------Run inference

And our cam_handler function where we print this last message

static void cam_handler(void *arg)
{
  uint32_t resolution = CAM_WIDTH * CAM_HEIGHT;
  uint32_t captureSize = resolution * sizeof(unsigned char);

  pi_camera_control(&camera, PI_CAMERA_CMD_STOP, 0);

  cpxPrintToConsole(LOG_TO_CRTP, "------Run inference\n");
  /* Run inference */
  pi_cluster_send_task_to_cl(&cluster_dev, task);
  
  cpxPrintToConsole(LOG_TO_CRTP, "------END inference\n");

  if (Output_1[0] > Output_1[1])
  {
    cpxPrintToConsole(LOG_TO_CRTP, "Cup,     confidence: %hd\n", Output_1[0] - Output_1[1]);
  }
  else
  {
    cpxPrintToConsole(LOG_TO_CRTP, "Background, confidence: %hd\n", Output_1[1] - Output_1[0]);
  }

  pi_camera_capture_async(&camera, cameraBuffer, CAM_WIDTH * CAM_HEIGHT, pi_task_callback(&task1, cam_handler, NULL));
  pi_camera_control(&camera, PI_CAMERA_CMD_START, 0); 

  if (wifiClientConnected == 1) {
    createImageHeaderPacket(&txp, captureSize, RAW_ENCODING);
    cpxSendPacketBlocking(&txp);
    sendBufferViaCPX(&txp, cameraBuffer, captureSize);
  }
}

Do you have any idea why our model doesn't work?

[hmllr]

Hi :) I would guess there is something in the task in which it gets stuck, so it is hard to help without this code - but this question anyway seems to be better suited for the "Discussions" (https://github.com/orgs/bitcraze/discussions) than as an issue, as it is about your custom model and not an issue in the example.