PCIe Tutorial

This tutorial walks you through using the PCIe Pipeline to enable high-performance video processing over the PCIe interface. You will learn how to run real-time object detection with YOLOv7 models on the SiMa.ai MLSoC platform.

Note

This section covers building PCIe pipelines with PePPi (Python).

Prerequisites:

Make sure sima_host_pcie_pkg.sh is installed on host for proper communication
Follow PCIe setup instructions to configure the PCIe cards on the host machine
PCIe pipeline is not supported on Modalix Early Access Kit and Modalix DevKit as they don’t have PCIe interface

Purpose

This application is designed to:

Ingest video input over PCIe from a host or connected capture device
Run YOLOv7-based object detection on each frame
Annotate frames with bounding boxes and labels
Stream the output back to the host or a connected interface (e.g., display, recording module)

This setup targets integration and benchmarking scenarios for PCIe-connected systems.

Configuration Overview

The runtime parameters are defined in project.yaml. The following tables provide a detailed breakdown of the configuration.

Input Configuration

Parameter	Description	Example
source.name	Input type used	`"pcie"`
source.value	Identifier string for PCIe input	`"PCIE"`
udp_host	Not used for PCIe setup	`""`
port	Not used for PCIe setup	`""`
pipeline	Inference pipeline identifier	`"YoloV7-Pcie"`

Model Configuration

Parameter	Description	Value
name	Model identifier	`"yolov7"`
targz	Path to the YOLOv7 model archive	`"<targz filepath>"`
label_file	Path to the label file	`"labels.txt"`
normalize	Enable input normalization	`true`
channel_mean	Per-channel mean values	`[0.0, 0.0, 0.0]`
channel_stddev	Per-channel stddev values	`[1.0, 1.0, 1.0]`
padding_type	Padding strategy during preprocessing	`"CENTER"`
aspect_ratio	Preserve input aspect ratio	`true`
topk	Maximum number of detections per frame	`10`
detection_threshold	Confidence threshold for object detection	`0.7`
decode_type	Postprocessing decode strategy	`"yolo"`

Project Configuration

project.yaml

source:
  name: "pcie"
  value: "PCIE"
udp_host: ""
port: ""
pipeline: "YoloV7-Pcie"

Models:
  - name: "yolov7"
    targz: "<targz filepath>"
    label_file: "labels.txt"
    normalize: true
    channel_mean: [0.0, 0.0, 0.0]
    channel_stddev: [1.0, 1.0, 1.0]
    padding_type: "CENTER"
    aspect_ratio: true
    topk: 10
    detection_threshold: 0.7
    decode_type: "yolo"

Script Behavior

The main script performs the following:

Loads the configuration from project.yaml
Initializes the PCIe VideoReader and VideoWriter
Sets the expected input and output resolution to 1280×720
Loads and configures the YOLOv7 model session on SiMa’s MLSoC
Continuously: - Reads frames from PCIe input - Runs detection inference - Annotates the output frame - Writes the result using PCIe streaming with metadata passthrough

Warning

Make sure the PCIe input source delivers frames at 1280x720 resolution, or update the script accordingly.

Model Details

Property	Details
Model File	Tiny YoloV7 optimzied for SiMa devices
Model Type	YOLOv7 Tiny
Normalization	Yes (mean: `[0.0, 0.0, 0.0]`, stddev: `[1.0, 1.0, 1.0]`)
Detection Threshold	0.7
Output	Top 10 detections per frame
Decoding Strategy	YOLO

Host Pipeline Setup

Run the Host side pipeline with the following command:

GST_DEBUG=3 gst-launch-1.0 -v rtspsrc location=<rtsp_url> ! rtph264depay ! h264parse ! avdec_h264 ! videoconvert ! video/x-raw,format=NV12 ! pciehost queuedepth=30 fps=30 showfps=1 card-number=1 ! queue ! videoparse format=nv12 width=1280 height=720 ! videoconvert ! autovideosink

Build and Deploy

Follow the standard PePPi build and deploy process:

Prerequisite:

1. Make sure the device is connected via SDK for deployment of project to device. Please follow the setup device section and follow the same.

sima-user@docker-image-id:/home/docker/sima-cli/workspace/pcie_pipeline/YoloV7_pcie$ mpk device connect -t sima@192.168.1.20
ℹ Connecting to sima@192.168.1.20...
🔗 Connection established to 192.168.1.20 .
ℹ Fetching Device Plugin Version data file from : 192.168.1.20 ...
✔ Successfully fetched and updated Plugin Version Data file from : 192.168.1.20.

Prepare project directory:

sima-user@docker-image-id:/home/docker/sima-cli/workspace$ mkdir pcie_tutorial && cd pcie_tutorial
sima-user@docker-image-id:/home/docker/sima-cli/workspace/pcie_tutorial$ sima-cli install -v 1.7.0 samples/peppi-tutorials/pcie

Update project.yaml

source:
  name: "pcie"
  value: "PCIE"
udp_host: ""
port: ""
pipeline: "YoloV7-Pcie"

Models:
  - name: "yolov7"
    targz: "yolov7-tiny-opt_mpk.tar.gz"
    label_file: labels.txt
    normalize: true
    channel_mean: [0.0, 0.0, 0.0]
    channel_stddev: [1.0, 1.0, 1.0]
    padding_type: "CENTER"
    aspect_ratio: true
    topk: 10
    detection_threshold: 0.7
    decode_type: "yolo"

Create the pipeline package:

sima-user@docker-image-id:/home/docker/sima-cli/workspace/pcie_tutorial$ mpk create --peppi -s . -d . --main-file=main.py --yaml-file=project.yaml
ℹ Generating requirements.txt file...
✔ Generated requirements.txt.
ℹ Dowloading required packages...
✔ Dowloaded required packages.
ℹ Building Rpm...
✔ Rpm built successfully.
ℹ Creating mpk file...
✔ Mpk file created successfully at
/home/docker/sima-cli/workspace/pcie_tutorial/project.mpk

Deploy the pipeline package:

sima-user@docker-image-id:/home/docker/sima-cli/workspace/pcie_tutorial$ mpk deploy -f project.mpk
ℹ Checking if App YoloV7-Pcie Plugin Version Index File
/home/docker/sima-cli/workspace/pcie_tutorial/YoloV7-Pcie_plugin_version_index.json
exists...
❗ App YoloV7-Pcie : File doesn't exist at
/home/docker/sima-cli/workspace/pcie_tutorial/YoloV7-Pcie_plugin_version_index.json.
Please check the path/file.
❔ Proceed by Skipping Plugin Version Check? [y/n]: y
‼ User chose to proceed with App YoloV7-Pcie deployment, Skipping Plugin Version Check! Performance
could be sub-optimal!
🚀 Sending MPK to 192.168.1.20...
Transfer Progress for project.mpk:  100.00%
🏁 MPK sent successfully!
Installing MPK... ━━━━━━━━━━╺━━━━━━━━━━━━━━━━━━━━━━━━━━━━━  25%

... ... ...

🏁 MPK sent successfully!
✔ MPK Deployed! ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100%
✔ MPK Deployment is successful for project.mpk.

Visualize the results:

On the host machine run the following commands card-number parameter in pciehost plugin needs to be set accordingly. 0 if only 1 card is available else “i” based on the card to process the input frames/data.

GST_DEBUG=3 gst-launch-1.0 -v rtspsrc location=<rtsp_url> ! rtph264depay ! h264parse ! avdec_h264 ! videoconvert ! video/x-raw,format=NV12 ! pciehost queuedepth=30 fps=30 showfps=1 card-number=1 ! queue ! videoparse format=nv12 width=1280 height=720 ! videoconvert ! autovideosink

To create a rtsp stream - Use the below commands:

1. docker run --rm -it --network=host bluenviron/mediamt
2. ffmpeg -re -stream_loop -1 -i <input_video_feed>.mp4 -c copy -f rtsp rtsp://<host_ip>:8554/mystream