Xilinx Kria KV260 Workshop

https://github.com/xilinx/xilinx_kria_kv260_workshop
Admin
A comprehensive workshop for setting up the Xilinx KV260 Starter kit board and running Accelerated...
Tokens:13,286
Snippets:121
Trust Score:9.3
License:Apache-2.0
Update:6 months ago
Show doc for...
Tokens:
Raw
### Example Caffe Model Compilation

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 4: Optional Demo.md

A concrete example of compiling the ssd_pedestrian_pruned_0_97 model using the Caffe compiler.

```bash
vai_c_caffe -p ssd_pedestrian_pruned_0_97/quantized/deploy.prototxt -c ssd_pedestrian_pruned_0_97/quantized/deploy.caffemodel -a arch.json -o compile_model -n ssd_pedestrian_pruned_0_97
```

--------------------------------

### Launch Jupyter Lab on SOM

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

Commands to start the Jupyter Lab server on the SOM and retrieve the access URL for browser-based interaction.

```bash
jupyter-lab --ip=ip-address &
sudo jupyter notebook list
```

--------------------------------

### Install Remmina for Remote Desktop Access (Linux/Mac)

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/How to connect to AWS.md

This command installs the Remmina remote desktop client on Ubuntu-based systems, enabling users to connect to the AWS instance via RDP. Ensure you have sudo privileges to install packages.

```bash
sudo apt install remmina
```

--------------------------------

### Launch Remmina Remote Desktop Client (Linux/Mac)

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/How to connect to AWS.md

This command launches the Remmina application after it has been installed. Users can then configure and initiate an RDP connection to the AWS instance.

```bash
remmina
```

--------------------------------

### Analyze Latency with GstShark

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

Steps to install GstShark, run the smart camera with latency tracing enabled, and extract inter-latency metrics from the generated log file.

```bash
dnf search GstShark
sudo dnf install libgstshark0.aarch64
sudo GST_DEBUG="GST_TRACER:7" GST_TRACERS=interlatency smartcam --file ${file.h264} -i h264 -W 1920 -H 1080 -r 30 --target dp --aitask facedetect >& facedetect-log.txt &
grep -rn interlatency facedetect-log.txt | tail -n 20 | cut -d'=' -f2-4
```

--------------------------------

### Application Management with xmutil

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

Commands for managing accelerated applications on the KV260 board, including listing, installing, loading, and unloading applications.

```APIDOC
## Application Management with xmutil

### Description
The `xmutil` utility manages accelerated applications on the KV260 board, allowing users to list available applications, install packages, and load/unload application firmware.

### List Available Applications
```bash
# List all available accelerated applications on the board

xmutil listapps
```

### Get Available Packages
```bash
# Get list of available packages to install

xmutil getpkgs
```

### Install Packages
```bash
# Install AA1 Smart Camera package

sudo dnf install packagegroup-kv260-smartcam.noarch

# Install AA2 AI Box package

sudo dnf install packagegroup-kv260-aibox-reid.noarch

# Install AA4 Defect Detection package

sudo dnf install packagegroup-kv260-defect-detect.noarch
```

### Load/Unload Applications
```bash
# Unload any currently running application

sudo xmutil unloadapp

# Load Smart Camera application (AA1)

sudo xmutil loadapp kv260-smartcam
# Output: Loaded kv260-smartcam successfully

# Load AI Box application (AA2)

sudo xmutil loadapp kv260-aibox-reid

# Unload specific application when done

xmutil unloadapp kv260-aa1
```
```

--------------------------------

### Manage Applications with xmutil

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

The `xmutil` utility is used to manage accelerated applications on the KV260 board. It allows listing, installing, loading, and unloading application packages and firmware. Dependencies include `sudo` for installation and loading/unloading operations.

```bash
xmutil listapps

xmutil getpkgs

sudo dnf install packagegroup-kv260-smartcam.noarch

sudo dnf install packagegroup-kv260-aibox-reid.noarch

sudo dnf install packagegroup-kv260-defect-detect.noarch

sudo xmutil unloadapp

sudo xmutil loadapp kv260-smartcam
# Output: Loaded kv260-smartcam successfully

sudo xmutil loadapp kv260-aibox-reid

xmutil unloadapp kv260-aa1
```

--------------------------------

### Deploy Custom AI Models with Vitis-AI on KV260

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

Guides users through deploying custom AI models from the Xilinx Model Zoo to the KV260 using Vitis-AI. Covers launching the Docker container, activating framework environments, compiling models (Caffe, TensorFlow, PyTorch), and copying them to the board.

```bash
# Launch Vitis-AI Docker container
cd Vitis-AI
./docker_run.sh xilinx/vitis-ai-cpu:latest

# Activate appropriate framework environment
conda activate vitis-ai-caffe      # For Caffe models
conda activate vitis-ai-tensorflow  # For TensorFlow 1.15
conda activate vitis-ai-tensorflow2 # For TensorFlow 2.3
conda activate vitis-ai-pytorch     # For PyTorch

# Create arch.json with KV260 fingerprint
cat > arch.json << 'EOF'
{
    "fingerprint":"0x1000020F6014406"
}
EOF

# Compile Caffe model for KV260
vai_c_caffe -p ssd_pedestrian_pruned_0_97/quantized/deploy.prototxt \
            -c ssd_pedestrian_pruned_0_97/quantized/deploy.caffemodel \
            -a arch.json -o compile_model -n ssd_pedestrian_pruned_0_97

# Compile TensorFlow model
vai_c_tensorflow -f ./quantize_eval_model.pb -a arch.json -o ./output -n mymodel

# Compile PyTorch model
vai_c_xir -x ./quantized.xmodel -a arch.json -o ./output -n mymodel

# Copy compiled model files to KV260 board
scp -r ./compile_model/ssd_pedestrian_pruned_0_97 root@<board-ip>:/opt/xilinx/share/vitis_ai_library/models/kv260-smartcam/
```

--------------------------------

### Play RTSP Stream from KV260 on Client Machines

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

Provides instructions for viewing RTSP streams originating from the KV260 on client machines using `ffplay` on Linux or media players like VLC on Windows/Mac. Includes steps to install `ffplay` if necessary.

```bash
# Play RTSP stream on Linux client
ffplay rtsp://192.168.1.26:5000/test

# Install ffplay on Linux if not available
sudo apt install ffmpeg
```

--------------------------------

### Start RTSP Stream for MIPI Camera

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

Configures the Smartcam application to capture video from the MIPI camera and stream it via an RTSP server. Users can specify custom resolution using width and height flags.

```bash
smartcam --mipi -t rtsp -W 1920 -H 1080
```

--------------------------------

### Exit and List Applications using xmutil

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

These commands are used to manage the AA1 application and other installed applications on the Kria KV260. `xmutil unloadapp kv260-aa1` exits the current application, while `xmutil listapps` displays all available applications.

```bash
xmutil unloadapp kv260-aa1
```

```bash
xmutil listapps
```

--------------------------------

### View RTSP Stream with FFplay

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

Uses the FFplay utility to connect to and display an active RTSP video stream. This command is compatible with Linux, Windows, and macOS environments provided FFmpeg is installed.

```bash
ffplay rtsp://<ip address>:554/test
```

--------------------------------

### Run Defect Detection Demo and Playback Modes

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

Demonstrates how to run the defect detection application in demo mode for live analysis or file playback mode for offline processing. Includes options for resolution, frame rate, and input/output files.

```bash
# Demo mode for demonstration purposes
defect-detect -w 1280 -h 800 -r 60 -f 0 -d 1 -m /dev/media0

# File playback mode with multiple output files
defect-detect -w 1280 -h 800 -r 60 -f 1 -i input.yuv -x raw.yuv -y pre_pros.yuv -z final.yuv

# Stop application with CTRL-C
```

--------------------------------

### Launch and Access Jupyter Lab Server on KV260

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

Instructions for launching a Jupyter Lab server on the KV260 board and accessing interactive notebooks via a web browser. Includes commands to find the board's IP address and list running notebook URLs.

```bash
# Launch Jupyter Lab server (get IP from ifconfig)
jupyter-lab --ip=192.168.1.26 &

# List running notebook URLs
sudo jupyter notebook list
```

--------------------------------

### Run Defect Detection Application

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 2: Exploring the Different AAs.md

Commands to load the defect detection application and execute it in normal, demo, or file playback modes.

```bash
sudo xmutil unloadapp
xmutil loadapp kv260-aa4
defect-detect -w 1280 -h 800 -r 60 -f 0 -d 0 -m /dev/media0
defect-detect -w 1280 -h 800 -r 60 -f 0 -d 1 -m /dev/media0
defect-detect -w 1280 -h 800 -r 60 -f 1 -i input.yuv -x raw.yuv -y pre_pros.yuv -z final.yuv
```

--------------------------------

### Troubleshoot Hardware and Applications

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/FAQ.md

Common command-line utilities for diagnosing hardware connectivity and application status on the Kria KV260 board.

```bash
ls /dev/media*
sudo xmutil unloadapp
xmutil loadapp kv260-smartcam
modetest -M xlnx
```

--------------------------------

### Initialize Vitis AI Docker Environment

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 4: Optional Demo.md

Commands to navigate to the Vitis AI directory and launch the Docker container for model compilation.

```bash
cd Vitis-AI
./docker_run.sh xilinx/vitis-ai-cpu:latest
```

--------------------------------

### Load AI Application and Configure Display

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 2: Exploring the Different AAs.md

Commands to unload existing applications, load the AI box application, and configure the display output for 4K resolution using modetest.

```bash
sudo xmutil unloadapp
sudo xmutil loadapp kv260-aibox-reid
sudo modetest -M xlnx -D 80000000.v_mix -s 52@40:3840x2160@NV16
```

--------------------------------

### Manage SOM Platform with xmutil

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/FAQ.md

The xmutil utility is used to manage the Kria SOM platform, including loading/unloading applications, checking performance, and updating firmware. These commands are executed on the target hardware command line interface.

```bash
xmutil boardid
xmutil bootfw
xmutil bootfwupd
xmutil getpkgs
xmutil listapps
xmutil loadapp <app_name>
xmutil unloadapp
xmutil perfmon
xmutil ddrqos
xmutil axiqos
xmutil bist
```

--------------------------------

### Stream Single Channel Video

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 2: Exploring the Different AAs.md

Commands to initialize a video stream using smartcam and observe the output with the aibox-reid application.

```bash
sudo smartcam -f ${file.h264} -W 1920 -H 1080 -r 30 -t rtsp -p 5000 -n &
sudo aibox-reid -s rtsp://192.168.1.26:5000/test -t rtsp -p 0
```

--------------------------------

### Output Smartcam Detections to File

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

This command demonstrates mixing input and output configurations for the smartcam application. It takes input from a MIPI sensor and directs the detection output to a file, suppressing standard output and error messages.

```bash
smartcam --mipi -W 1920 -H 1080 --target file >/dev/null 2>&1
```

--------------------------------

### Measure Smart Camera Performance

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

Commands to execute the smart camera application with specific parameters to measure throughput (FPS) and power consumption.

```bash
sudo smartcam --file ${file.h264} -i h264 -r 30 --target dp --aitask facedetect -R
xmutil platformstats -p
sudo smartcam --file ${file.h264} -i h264 -W 1920 -H 1080 -r 30 --target dp --aitask facedetect &
```

--------------------------------

### Run Smart Camera (AA1) Application

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

The `smartcam` command executes AI tasks such as face detection, pedestrian detection, and ADAS object detection using hardware acceleration. It supports various input sources (MIPI camera, USB camera, file) and output targets (RTSP, DisplayPort/HDMI, file). Options include specifying resolution, frame rate, AI task models, and display features.

```bash
# Basic MIPI camera input with 1080p resolution
sudo smartcam --mipi -W 1920 -H 1080

# USB camera input (device ID 1)
sudo smartcam --usb 1 -W 1920 -H 1080

# Stream MIPI camera to RTSP server
sudo smartcam --mipi -t rtsp -W 1920 -H 1080
# Output: stream ready at: rtsp://192.168.1.26:5000/test

# Stream to RTSP with audio input
sudo smartcam --mipi -W 1920 -H 1080 --target rtsp --audio

# Output to DisplayPort/HDMI monitor
sudo smartcam --mipi -W 1920 -H 1080 --target dp

# Save output to file
sudo smartcam --mipi -W 1920 -H 1080 --target file

# Process video file with face detection AI task
sudo smartcam --file ./video.h264 -i h264 -W 1920 -H 1080 -r 30 --target dp --aitask facedetect

# Use SSD ADAS model for car/bicycle/pedestrian detection
sudo smartcam --file ./traffic.h264 -i h264 -W 1920 -H 1080 -r 30 --target dp --aitask ssd

# Use RefineDet model
sudo smartcam --mipi -t rtsp -a refinedet

# File input to file output (generates out.h264)
sudo smartcam --file ./input.h264 --infile-type h264 --target file

# Display FPS on screen (note: causes performance degradation)
sudo smartcam --mipi -t rtsp -W 1920 -H 1080 -s

# Report FPS in terminal
sudo smartcam --file ./video.h264 -i h264 -r 30 --target dp --aitask facedetect -R

# Run as RTSP server without AI inference (for AA2 input source)
sudo smartcam -f ./video.h264 -W 1920 -H 1080 -r 30 -t rtsp -p 5000 -n &
# Output: stream ready at: rtsp://192.168.1.26:5000/test
```

--------------------------------

### Convert Video to H.264 and Transfer to KV260

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

Provides commands to convert video files to H.264 format using ffmpeg and transfer them to the KV260 board using SCP. Also suggests online converters for Windows/Mac users.

```bash
# Convert MP4 to H.264 using ffmpeg (Linux)
ffmpeg -i "input-video.mp4" -c:v libx264 -pix_fmt nv12 -r 30 output.nv12.h264

# Transfer file to board using SCP (from host machine)
scp ./output.nv12.h264 root@<board-ip>:/home/petalinux/
```

--------------------------------

### Measure AI Inference Performance on KV260

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

Shows how to measure Frames Per Second (FPS) throughput, monitor power consumption, and calculate end-to-end latency for AI inference applications on the KV260 using smartcam and GStreamer tools.

```bash
# Measure FPS throughput
sudo smartcam --file ./video.h264 -i h264 -r 30 --target dp --aitask facedetect -R

# Monitor power consumption in background
xmutil platformstats -p &
sudo smartcam --file ./video.h264 -i h264 -W 1920 -H 1080 -r 30 --target dp --aitask facedetect &

# Install GstShark for latency measurement
dnf search GstShark
sudo dnf install libgstshark0.aarch64

# Measure end-to-end latency with GStreamer pipeline
sudo GST_DEBUG="GST_TRACER:7" GST_TRACERS=interlatency smartcam --file ./video.h264 -i h264 \
    -W 1920 -H 1080 -r 30 --target dp --aitask facedetect >& facedetect-log.txt &

# Extract latency values from log
grep -rn interlatency facedetect-log.txt | tail -n 20 | cut -d'=' -f2-4
```

--------------------------------

### Display Frames Per Second (FPS) on Screen using Smartcam

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

This command configures the smartcam application to display the frames per second (FPS) on the screen. The `-s` flag is used to enable this feature, alongside specifying the video resolution and target output.

```bash
smartcam --mipi -t rtsp -W 1920 -H 1080 -s
```

--------------------------------

### Compile Models for Vitis AI

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 4: Optional Demo.md

Commands to compile quantized models into xmodel format using the Vitis AI compiler, tailored for different frameworks.

```bash
## For Caffe
vai_c_caffe -p </PATH/TO/deploy.prototxt> -c </PATH/TO/deploy.caffemodel> -a /<PATH/TO/arch.json> -o </OUTPUTPATH> -n <netname>

## For Tensorflow
vai_c_tensorflow -f </PATH/TO/quantize_eval_model.pb> <PATH/TO/arch.json> -o </OUTPUTPATH> -n <netname>

## Pytorch
vai_c_xir -x /PATH/TO/quantized.xmodel -a /PATH/TO/arch.json -o /OUTPUTPATH -n netname
```

--------------------------------

### Navigate Application Scripts

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

Commands to locate and list the pre-written shell scripts available for exploring AA1 features.

```bash
cd /opt/xilinx/bin
ls -ll
```

--------------------------------

### Run Defect Detection (AA4) Application

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

The `defect-detect` command is used for hardware-accelerated machine vision tasks specifically for manufacturing defect detection. It requires loading the appropriate application using `xmutil` and then running the `defect-detect` command with parameters for resolution, frame rate, input source, and media device.

```bash
# Load defect detection application
sudo xmutil unloadapp
sudo xmutil loadapp kv260-aa4

# Normal mode with live camera
defect-detect -w 1280 -h 800 -r 60 -f 0 -d 0 -m /dev/media0
```

--------------------------------

### Stream Video with Audio using Smartcam

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

This command streams video with audio capabilities enabled. It specifies the video resolution and the target output format (RTSP). The `--audio` flag is crucial for enabling audio streaming.

```bash
sudo smartcam --mipi -W 1920 -H 1080 --target rtsp --audio
```

--------------------------------

### Run AI Inference with RTSP Streaming

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

Executes the Smartcam application with a specific AI inference task (Refinedet) and streams the output over RTSP. Other supported tasks include facedetect and ssd.

```bash
sudo smartcam --mipi -t rtsp -a refindet
```

--------------------------------

### Write Image to SD Card using dd

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Linux set-up.md

This snippet demonstrates how to write a compressed image file (petalinux-sdimage.wic.gz) to an SD card using the `dd` command in Linux. It includes steps to decompress the image, write it to the specified device, and then safely eject the SD card. Ensure you identify the correct SD card device to avoid data loss.

```bash
gunzip petalinux-sdimage.wic.gz
dd if=petalinux-sdimage.wic of=/dev/sd<#> conv=fsync
sudo eject /dev/sd<#>
```

--------------------------------

### Run AI Box (AA2) - aibox-reid Application

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

The `aibox-reid` command performs multi-stream pedestrian detection and re-identification. It supports displaying up to 4 simultaneous 1080p video streams on a 4K monitor. Configuration involves setting the display mode and specifying input streams with their types and desired quadrant positions.

```bash
# Enable 4K display mode for multi-stream viewing
sudo modetest -M xlnx -D 80000000.v_mix -s 52@40:3840x2160@NV16

# Single channel RTSP input displayed in position 0 (top-left)
sudo aibox-reid -s rtsp://192.168.1.26:5000/test -t rtsp -p 0

# Two channels: RTSP stream (top-right) and file (bottom-left)
sudo aibox-reid -s rtsp://192.168.29.11:5000/test -t rtsp -p 2 \
                -s ./video.h264 -t file -p 1

# Four channels from video files in all quadrants
sudo aibox-reid -s ./video1.h264 -t file -p 0 \
                -s ./video2.h264 -t file -p 1 \
                -s ./video3.h264 -t file -p 2 \
                -s ./video4.h264 -t file -p 3

# Position options:
# -p 0: top-left
# -p 1: bottom-left
# -p 2: top-right
# -p 3: bottom-right
```

--------------------------------

### Process Video File to DP Monitor

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

Reads a local video file, applies the SSD object detection model, and displays the annotated output to a connected DisplayPort or HDMI monitor.

```bash
smartcam --file ${file.h264} --target dp -r 30 --aitask ssd
```

--------------------------------

### Identify Serial Port and Connect via Screen

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Mac set-up.md

Commands to list available serial devices on macOS and initiate a serial terminal session. The screen command requires the specific device path identified by the list command.

```bash
ls /dev/tty.*
screen /dev/tty.usbserial00022331
```

--------------------------------

### Process Video File to File

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 3: Running through AA1.md

Reads a local H.264 video file, performs detection, and saves the output to a new file. This is useful for offline processing and generating annotated video clips.

```bash
smartcam --file ${file.h264} --infile-type h264 --target file
```

--------------------------------

### Stream Multiple Video Channels

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 2: Exploring the Different AAs.md

Commands to process and display multiple video sources (RTSP and local files) simultaneously using the aibox-reid application.

```bash
sudo aibox-reid -s rtsp://192.168.29.11:5000/test -t rtsp -p 2 -s ${file.h264} -t file -p 1
sudo aibox-reid -s ${file1.h264} -t file -p 0 -s ${file2.h264} -t file -p 1 -s ${file3.h264} -t file -p 2 -s ${file4.h264} -t file -p 3
```

--------------------------------

### Configure AI Inference for Custom Models on KV260

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

Explains how to update the AI inference configuration file (`aiinference.json`) on the KV260 to use custom models that have been deployed. This involves specifying the model name, class, and path.

```json
{
    "ivas-library-repo": "/usr/lib/",
    "element-mode": "inplace",
    "kernels": [
        {
            "library-name": "libivas_xdpuinfer.so",
            "config": {
                "model-name": "ssd_pedestrian_pruned_0_97",
                "model-class": "REFINEDET",
                "model-path": "/opt/xilinx/share/vitis_ai_library/models/kv260-smartcam",
                "run_time_model": false,
                "need_preprocess": false,
                "performance_test": false,
                "debug_level": 0
            }
        }
    ]
}
```

--------------------------------

### Define Architecture Fingerprint

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 4: Optional Demo.md

JSON configuration file defining the hardware fingerprint required for model compilation on the KV260.

```json
{
    "fingerprint":"0x1000020F6014406"
}
```

--------------------------------

### Activate Vitis AI Framework Environments

Source: https://github.com/xilinx/xilinx_kria_kv260_workshop/blob/main/Part 4: Optional Demo.md

Commands to activate the specific Conda environment based on the model's source framework (Caffe, TensorFlow, or PyTorch).

```bash
conda activate vitis-ai-caffe
conda activate vitis-ai-tensorflow
conda activate vitis-ai-tensorflow2
conda activate vitis-ai-pytorch
```

--------------------------------

### Defect Detection (AA4) - defect-detect Command

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

Commands for running the Defect Detection application for manufacturing defect detection.

```APIDOC
## Defect Detection (AA4) - defect-detect Command

### Description
The `defect-detect` command runs hardware-accelerated machine vision for manufacturing defect detection applications.

### Usage
```bash
# Load defect detection application

sudo xmutil unloadapp
sudo xmutil loadapp kv260-aa4

# Normal mode with live camera

defect-detect -w 1280 -h 800 -r 60 -f 0 -d 0 -m /dev/media0
```
```

--------------------------------

### Smart Camera (AA1) - smartcam Command

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

Commands for running the Smart Camera application, which performs face detection, pedestrian detection, and ADAS object detection.

```APIDOC
## Smart Camera (AA1) - smartcam Command

### Description
The `smartcam` command runs face detection, pedestrian detection, and ADAS object detection using hardware-accelerated ML inference. It supports MIPI cameras, USB cameras, and file inputs with output to RTSP, DisplayPort/HDMI, or file.

### Basic Usage
```bash
# Basic MIPI camera input with 1080p resolution

sudo smartcam --mipi -W 1920 -H 1080

# USB camera input (device ID 1)

sudo smartcam --usb 1 -W 1920 -H 1080
```

### Output Options
```bash
# Stream MIPI camera to RTSP server

sudo smartcam --mipi -t rtsp -W 1920 -H 1080
# Output: stream ready at: rtsp://192.168.1.26:5000/test

# Stream to RTSP with audio input

sudo smartcam --mipi -W 1920 -H 1080 --target rtsp --audio

# Output to DisplayPort/HDMI monitor

sudo smartcam --mipi -W 1920 -H 1080 --target dp

# Save output to file

sudo smartcam --mipi -W 1920 -H 1080 --target file
```

### AI Task and Model Selection
```bash
# Process video file with face detection AI task

sudo smartcam --file ./video.h264 -i h264 -W 1920 -H 1080 -r 30 --target dp --aitask facedetect

# Use SSD ADAS model for car/bicycle/pedestrian detection

sudo smartcam --file ./traffic.h264 -i h264 -W 1920 -H 1080 -r 30 --target dp --aitask ssd

# Use RefineDet model

sudo smartcam --mipi -t rtsp -a refinedet
```

### File Input/Output
```bash
# File input to file output (generates out.h264)

sudo smartcam --file ./input.h264 --infile-type h264 --target file
```

### Performance Monitoring
```bash
# Display FPS on screen (note: causes performance degradation)

sudo smartcam --mipi -t rtsp -W 1920 -H 1080 -s

# Report FPS in terminal

sudo smartcam --file ./video.h264 -i h264 -r 30 --target dp --aitask facedetect -R
```

### RTSP Server without AI Inference
```bash
# Run as RTSP server without AI inference (for AA2 input source)

sudo smartcam -f ./video.h264 -W 1920 -H 1080 -r 30 -t rtsp -p 5000 -n &
# Output: stream ready at: rtsp://192.168.1.26:5000/test
```
```

--------------------------------

### AI Box (AA2) - aibox-reid Command

Source: https://context7.com/xilinx/xilinx_kria_kv260_workshop/llms.txt

Commands for running the AI Box application, which performs multi-stream pedestrian detection and re-identification.

```APIDOC
## AI Box (AA2) - aibox-reid Command

### Description
The `aibox-reid` command performs multi-stream pedestrian detection and re-identification on up to 4 simultaneous 1080p video streams displayed on a 4K monitor.

### Enable 4K Display Mode
```bash
# Enable 4K display mode for multi-stream viewing

sudo modetest -M xlnx -D 80000000.v_mix -s 52@40:3840x2160@NV16
```

### Multi-Stream Configuration
```bash
# Single channel RTSP input displayed in position 0 (top-left)

sudo aibox-reid -s rtsp://192.168.1.26:5000/test -t rtsp -p 0

# Two channels: RTSP stream (top-right) and file (bottom-left)

sudo aibox-reid -s rtsp://192.168.29.11:5000/test -t rtsp -p 2 \
                -s ./video.h264 -t file -p 1

# Four channels from video files in all quadrants

sudo aibox-reid -s ./video1.h264 -t file -p 0 \
                -s ./video2.h264 -t file -p 1 \
                -s ./video3.h264 -t file -p 2 \
                -s ./video4.h264 -t file -p 3
```

### Position Options
```
# -p 0: top-left
# -p 1: bottom-left
# -p 2: top-right
# -p 3: bottom-right
```
```
Tokens:
Raw