GstInference GStreamer pipelines for Jetson TX2

From RidgeRun Developer Wiki




Previous: Example pipelines/NANO Index Next: Example pipelines/Xavier






Problems running the pipelines shown on this page? Please see our GStreamer Debugging guide for help.

Tensorflow

Inceptionv4 inference on image file using Tensorflow

IMAGE_FILE=cat.jpg
MODEL_LOCATION='graph_inceptionv4_tensorflow.pb'
INPUT_LAYER='input'
OUTPUT_LAYER='InceptionV4/Logits/Predictions'
GST_DEBUG=inceptionv4:6 gst-launch-1.0 \
multifilesrc location=$IMAGE_FILE ! jpegparse ! nvjpegdec ! 'video/x-raw' ! nvvidconv ! 'video/x-raw(memory:NVMM),format=NV12' ! nvvidconv ! queue ! net.sink_model \
inceptionv4 name=net model-location=$MODEL_LOCATION backend=tensorflow backend::input-layer=$INPUT_LAYER  backend::output-layer=$OUTPUT_LAYER -
  • Output
0:02:22.005527960 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:02:22.168796723 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:02:22.168947603 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.627314)
0:02:22.169237202 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:02:22.339393463 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:02:22.339496918 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.627314)
0:02:22.339701878 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:02:22.507804674 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:02:22.507950081 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.627314)
0:02:22.508232128 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:02:22.678740356 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:02:22.678892356 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.627314)

Inceptionv4 inference on video file using TensorFlow

VIDEO_FILE='cat.mp4'
MODEL_LOCATION='graph_inceptionv4_tensorflow.pb'
INPUT_LAYER='input'
OUTPUT_LAYER='InceptionV4/Logits/Predictions'
GST_DEBUG=inceptionv4:6 gst-launch-1.0 \
filesrc location=$VIDEO_FILE ! qtdemux name=demux ! h264parse ! omxh264dec ! nvvidconv ! queue ! net.sink_model \
inceptionv4 name=net model-location=$MODEL_LOCATION backend=tensorflow backend::input-layer=$INPUT_LAYER  backend::output-layer=$OUTPUT_LAYER
  • Output
0:00:11.728307018 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:11.892030154 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:11.892258185 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.686857)
0:00:11.892556808 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.065318539 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.065467786 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.673300)
0:00:12.065759849 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.247159695 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.247309295 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.669102)
0:00:12.247612718 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.419172436 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.419321396 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.667991)

Inceptionv4 inference on camera stream using TensorFlow

  • Get the graph used on this example from this link
  • You will need a camera compatible with NVIDIA Libargus API or V4l2.

NVIDIA Camera

  • Pipeline
SENSOR_ID=0
MODEL_LOCATION='graph_inceptionv4_tensorflow.pb'
INPUT_LAYER='input'
OUTPUT_LAYER='InceptionV4/Logits/Predictions'
GST_DEBUG=inceptionv4:6 gst-launch-1.0 \
nvcamerasrc sensor-id=$SENSOR_ID ! nvvidconv ! queue ! net.sink_model \
inceptionv4 name=net backend=tensorflow model-location=$MODEL_LOCATION backend::input-layer=$INPUT_LAYER backend::output-layer=$OUTPUT_LAYER

V4L2

  • Pipeline
CAMERA='/dev/video1'
MODEL_LOCATION='graph_inceptionv4_tensorflow.pb'
INPUT_LAYER='input'
OUTPUT_LAYER='InceptionV4/Logits/Predictions'
GST_DEBUG=inceptionv4:6 gst-launch-1.0 \
v4l2src device=$CAMERA ! videoconvert ! videoscale ! queue ! net.sink_model \
inceptionv4 name=net model-location=$MODEL_LOCATION backend=tensorflow backend::input-layer=$INPUT_LAYER  backend::output-layer=$OUTPUT_LAYER
  • Output
0:00:12.199657219  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.365172092  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.365271548  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 774 : (0.196048)
0:00:12.365421435  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.530604726  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.530700501  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 774 : (0.179406)
0:00:12.530848565  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.697053611  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.697147818  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 774 : (0.144033)
0:00:12.697295530  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.862007878  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.862104134  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 774 : (0.157707)
0:00:12.862252645  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:13.027090881  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:13.027190273  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 774 : (0.142998)

Inceptionv4 visualization with classification overlay Tensorflow

  • Get the graph used on this example from this link
  • You will need a camera compatible with NVIDIA Libargus API or V4l2.

NVIDIA Camera

  • Pipeline
SENSOR_ID=0
MODEL_LOCATION='graph_inceptionv4_tensorflow.pb'
INPUT_LAYER='input'
OUTPUT_LAYER='InceptionV4/Logits/Predictions'
LABELS='imagenet_labels.txt'
gst-launch-1.0 \
nvcamerasrc sensor-id=$SENSOR_ID ! 'video/x-raw(memory:NVMM)' ! tee name=t \
t. ! queue max-size-buffers=1 leaky=downstream ! nvvidconv ! 'video/x-raw,format=(string)RGBA' ! net.sink_model \
t. ! queue max-size-buffers=1 leaky=downstream ! nvvidconv ! 'video/x-raw,format=(string)RGBA' ! net.sink_bypass \
inceptionv4 name=net backend=tensorflow model-location=$MODEL_LOCATION backend::input-layer=$INPUT_LAYER backend::output-layer=$OUTPUT_LAYER \
net.src_bypass ! classificationoverlay labels="$(cat $LABELS)" font-scale=4 thickness=4 ! nvvidconv ! nvoverlaysink sync=false -v

V4L2

  • Pipeline
CAMERA='/dev/video1'
MODEL_LOCATION='graph_inceptionv4_tensorflow.pb'
INPUT_LAYER='input'
OUTPUT_LAYER='InceptionV4/Logits/Predictions'
LABELS='imagenet_labels.txt'
gst-launch-1.0 \
v4l2src device=$CAMERA ! "video/x-raw, width=1280, height=720" ! tee name=t \
t. ! videoconvert ! videoscale ! queue ! net.sink_model \
t. ! queue ! net.sink_bypass \
inceptionv4 name=net model-location=$MODEL_LOCATION backend=tensorflow backend::input-layer=$INPUT_LAYER  backend::output-layer=$OUTPUT_LAYER \
net.src_bypass ! videoconvert ! classificationoverlay labels="$(cat $LABELS)" font-scale=4 thickness=4 ! videoconvert ! xvimagesink sync=false
  • Output
Example classification overlay output

TinyYolov2 inference on image file using Tensorflow

  • Get the graph used on this example from this link
  • You will need an image file from one of TinyYOLO classes
  • Pipeline
IMAGE_FILE='cat.jpg'
MODEL_LOCATION='graph_tinyyolov2_tensorflow.pb'
INPUT_LAYER='input/Placeholder'
OUTPUT_LAYER='add_8'
GST_DEBUG=tinyyolov2:6 gst-launch-1.0 \
multifilesrc location=$IMAGE_FILE ! jpegparse ! nvjpegdec ! 'video/x-raw' ! nvvidconv ! 'video/x-raw(memory:NVMM),format=NV12' ! nvvidconv ! queue ! net.sink_model \
tinyyolov2 name=net model-location=$MODEL_LOCATION backend=tensorflow backend::input-layer=$INPUT_LAYER  backend::output-layer=$OUTPUT_LAYER
  • Output
0:00:07.137677204 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.266928985 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.267080761 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:25.820670, y:11.977936, width:425.495203, height:450.224357, prob:15.204609]
0:00:07.267382968 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.394225925 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.394431653 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:25.820670, y:11.977936, width:425.495203, height:450.224357, prob:15.204609]
0:00:07.394858915 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.527547133 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.527753020 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:25.820670, y:11.977936, width:425.495203, height:450.224357, prob:15.204609]
0:00:07.528080219 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.662473455 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.662769998 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:25.820670, y:11.977936, width:425.495203, height:450.224357, prob:15.204609]

TinyYolov2 inference on video file using Tensorflow

  • Get the graph used on this example from this link
  • You will need a video file from one of TinyYOLO classes
  • Pipeline
VIDEO_FILE='cat.mp4'
MODEL_LOCATION='graph_tinyyolov2_tensorflow.pb'
INPUT_LAYER='input/Placeholder'
OUTPUT_LAYER='add_8'
GST_DEBUG=tinyyolov2:6 gst-launch-1.0 \
filesrc location=$VIDEO_FILE ! qtdemux name=demux ! h264parse ! omxh264dec ! nvvidconv ! queue ! net.sink_model \
tinyyolov2 name=net model-location=$MODEL_LOCATION backend=tensorflow backend::input-layer=$INPUT_LAYER  backend::output-layer=$OUTPUT_LAYER
  • Output
0:00:07.245722660 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.360377432 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.360586455 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:-46.105452, y:-9.139365, width:445.139551, height:487.967720, prob:14.592537]
0:00:07.360859318 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.489190714 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.489382873 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:-46.140270, y:-9.193503, width:445.228762, height:488.028163, prob:14.596972]
0:00:07.489736216 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.629190069 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.629379733 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:-46.281640, y:-9.164348, width:445.512899, height:487.908826, prob:14.596945]
0:00:07.629717876 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.761072493 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.761271244 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:-46.338202, y:-9.202273, width:445.624841, height:487.954952, prob:14.592540]

TinyYolov2 inference on camera stream using Tensorflow

  • Get the graph used on this example from this link
  • You will need a camera compatible with NVIDIA Libargus API or V4l2.

NVIDIA Camera

  • Pipeline
SENSOR_ID=0
MODEL_LOCATION='graph_tinyyolov2_tensorflow.pb'
INPUT_LAYER='input/Placeholder'
OUTPUT_LAYER='add_8'
GST_DEBUG=tinyyolov2:6 gst-launch-1.0 \
nvarguscamerasrc sensor-id=$SENSOR_ID ! nvvidconv ! 'video/x-raw,format=BGRx' ! queue ! net.sink_model \
tinyyolov2 name=net model-location=$MODEL_LOCATION backend=tensorflow backend::input-layer=$INPUT_LAYER  backend::output-layer=$OUTPUT_LAYER

V4L2

  • Pipeline
CAMERA='/dev/video1'
MODEL_LOCATION='graph_tinyyolov2_tensorflow.pb'
INPUT_LAYER='input/Placeholder'
OUTPUT_LAYER='add_8'
GST_DEBUG=tinyyolov2:6 gst-launch-1.0 \
v4l2src device=$CAMERA ! videoconvert ! videoscale ! queue ! net.sink_model \
tinyyolov2 name=net model-location=$MODEL_LOCATION backend=tensorflow backend::input-layer=$INPUT_LAYER  backend::output-layer=$OUTPUT_LAYER
  • Output
0:00:39.754924355  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:39.876816786  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:39.876914225  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:4, x:147.260736, y:116.184709, width:134.389472, height:245.113627, prob:8.375733]
0:00:39.877085489  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:39.999699614  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:39.999799198  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:4, x:146.957935, y:117.902112, width:134.883825, height:242.143126, prob:7.982772]
0:00:39.999962206  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:40.118613969  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:40.118712017  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:4, x:147.147349, y:116.562615, width:134.469630, height:244.181931, prob:8.139100]
0:00:40.118882641  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:40.264861052  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:40.264964828  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:4, x:146.618516, y:117.162739, width:135.454029, height:243.785573, prob:8.112847]

TinyYolov2 visualization with detection overlay Tensorflow

  • Get the graph used on this example from this link
  • You will need a camera compatible with NVIDIA Libargus API or V4l2.

NVIDIA Camera

  • Pipeline
SENSOR_ID=0
MODEL_LOCATION='graph_tinyyolov2_tensorflow.pb'
INPUT_LAYER='input/Placeholder'
OUTPUT_LAYER='add_8'
LABELS='labels.txt'
GST_DEBUG=tinyyolov2:6 \
gst-launch-1.0 \
nvcamerasrc sensor-id=$SENSOR_ID ! 'video/x-raw(memory:NVMM)' ! tee name=t \
t. ! queue max-size-buffers=1 leaky=downstream ! nvvidconv ! 'video/x-raw,format=(string)RGBA' ! net.sink_model \
t. ! queue max-size-buffers=1 leaky=downstream ! nvvidconv ! 'video/x-raw,format=(string)RGBA' ! net.sink_bypass \
tinyyolov2 name=net backend=tensorflow model-location=$MODEL_LOCATION backend::input-layer=$INPUT_LAYER backend::output-layer=$OUTPUT_LAYER \
net.src_bypass !  detectionoverlay labels="$(cat $LABELS)" font-scale=4 thickness=4  ! nvvidconv ! nvoverlaysink sync=false -v

V4L2

  • Pipeline
CAMERA='/dev/video1'
MODEL_LOCATION='graph_tinyyolov2_tensorflow.pb'
INPUT_LAYER='input/Placeholder'
OUTPUT_LAYER='add_8'
LABELS='labels.txt'
gst-launch-1.0 \
v4l2src device=$CAMERA ! "video/x-raw, width=1280, height=720" ! tee name=t \
t. ! videoconvert ! videoscale ! queue ! net.sink_model \
t. ! queue ! net.sink_bypass \
tinyyolov2 name=net model-location=$MODEL_LOCATION backend=tensorflow backend::input-layer=$INPUT_LAYER backend::output-layer=$OUTPUT_LAYER \
net.src_bypass ! videoconvert ! detectionoverlay labels="$(cat $LABELS)" font-scale=1 thickness=2 ! videoconvert ! xvimagesink sync=false
  • Output
Example detection overlay output

FaceNet visualization with embedding overlay Tensorflow

  • Get the graph used on this example from this link
  • You will need a camera compatible with NVIDIA Libargus API or V4l2.
  • LABELS and EMBEDDINGS files are in $PATH_TO_GST_INFERENCE_ROOT_DIR/tests/examples/embedding/embeddings.

NVIDIA Camera

  • Pipeline
SENSOR_ID=0
MODEL_LOCATION='graph_facenetv1_tensorflow.pb'
INPUT_LAYER='input'
OUTPUT_LAYER='output'
LABELS='$PATH_TO_GST_INFERENCE_ROOT_DIR/tests/examples/embedding/embeddings/labels.txt'
EMBEDDINGS='$PATH_TO_GST_INFERENCE_ROOT_DIR/tests/examples/embedding/embeddings/embeddings.txt'
gst-launch-1.0 \
nvcamerasrc sensor-id=$SENSOR_ID ! 'video/x-raw(memory:NVMM),width=(int)1280,height=(int)720' ! nvvidconv ! 'video/x-raw,format=BGRx,width=(int)1280,height=(int)720' ! videoconvert ! tee name=t \
t. ! queue ! videoscale ! queue ! net.sink_model \
t. ! queue ! net.sink_bypass \
facenetv1 name=net model-location=$MODEL_LOCATION backend=tensorflow backend::input-layer=$INPUT_LAYER backend::output-layer=$OUTPUT_LAYER \
net.src_bypass ! videoconvert ! embeddingoverlay labels="$(cat $LABELS)" embeddings="$(cat $EMBEDDINGS)" font-scale=4 thickness=4 ! videoconvert ! xvimagesink sync=false

V4L2

  • Pipeline
CAMERA='/dev/video1'
MODEL_LOCATION='graph_facenetv1_tensorflow.pb'
INPUT_LAYER='input'
OUTPUT_LAYER='output'
LABELS='$PATH_TO_GST_INFERENCE_ROOT_DIR/tests/examples/embedding/embeddings/labels.txt'
EMBEDDINGS='$PATH_TO_GST_INFERENCE_ROOT_DIR/tests/examples/embedding/embeddings/embeddings.txt'
gst-launch-1.0 \
v4l2src device=$CAMERA ! "video/x-raw, width=1280, height=720" ! tee name=t \
t. ! videoconvert ! videoscale ! queue ! net.sink_model \
t. ! queue ! net.sink_bypass \
facenetv1 name=net model-location=$MODEL_LOCATION backend=tensorflow backend::input-layer=$INPUT_LAYER backend::output-layer=$OUTPUT_LAYER \
net.src_bypass ! videoconvert ! embeddingoverlay labels="$(cat $LABELS)" embeddings="$(cat $EMBEDDINGS)" font-scale=4 thickness=4 ! videoconvert ! xvimagesink sync=false


  • Output
Example embedding overlay output

TensorFlow-Lite

Inceptionv4 inference on image file using TensorFlow-Lite

IMAGE_FILE='cat.jpg'
MODEL_LOCATION='graph_inceptionv4.tflite'
LABELS='labels.txt'
GST_DEBUG=inceptionv4:6 gst-launch-1.0 \
multifilesrc location=$IMAGE_FILE ! jpegparse ! nvjpegdec ! 'video/x-raw' ! nvvidconv ! 'video/x-raw(memory:NVMM),format=NV12' ! nvvidconv ! queue ! net.sink_model \
inceptionv4 name=net model-location=$MODEL_LOCATION backend=tflite labels="$(cat $LABELS)"
  • Output
0:02:22.005527960 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:02:22.168796723 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:02:22.168947603 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.627314)
0:02:22.169237202 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:02:22.339393463 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:02:22.339496918 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.627314)
0:02:22.339701878 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:02:22.507804674 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:02:22.507950081 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.627314)
0:02:22.508232128 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:02:22.678740356 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:02:22.678892356 30355       0x5accf0 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.627314)

Inceptionv4 inference on video file using TensorFlow-Lite

VIDEO_FILE='cat.mp4'
MODEL_LOCATION='graph_inceptionv4.tflite'
LABELS='labels.txt'
GST_DEBUG=inceptionv4:6 gst-launch-1.0 \
filesrc location=$VIDEO_FILE ! qtdemux name=demux ! h264parse ! omxh264dec ! nvvidconv ! queue ! net.sink_model \
inceptionv4 name=net model-location=$MODEL_LOCATION backend=tflite labels="$(cat $LABELS)"
  • Output
0:00:11.728307018 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:11.892030154 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:11.892258185 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.686857)
0:00:11.892556808 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.065318539 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.065467786 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.673300)
0:00:12.065759849 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.247159695 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.247309295 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.669102)
0:00:12.247612718 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.419172436 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.419321396 30399       0x5ad000 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 282 : (0.667991)

Inceptionv4 inference on camera stream using TensorFlow-Lite

  • Get the graph used on this example from this link
  • You will need a camera compatible with NVIDIA Libargus API or V4l2.

NVIDIA Camera

  • Pipeline
SENSOR_ID=0
MODEL_LOCATION='graph_inceptionv4.tflite'
LABELS='labels.txt'
GST_DEBUG=inceptionv4:6 gst-launch-1.0 \
nvcamerasrc sensor-id=$SENSOR_ID ! nvvidconv ! queue ! net.sink_model \
inceptionv4 name=net backend=tflite model-location=$MODEL_LOCATION labels="$(cat $LABELS)"

V4L2

  • Pipeline
CAMERA='/dev/video0'
MODEL_LOCATION='graph_inceptionv4.tflite'
LABELS='labels.txt'
GST_DEBUG=inceptionv4:6 gst-launch-1.0 \
v4l2src device=$CAMERA ! videoconvert ! videoscale ! queue ! net.sink_model \
inceptionv4 name=net model-location=$MODEL_LOCATION backend=tflite labels="$(cat $LABELS)"
  • Output
0:00:12.199657219  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.365172092  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.365271548  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 774 : (0.196048)
0:00:12.365421435  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.530604726  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.530700501  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 774 : (0.179406)
0:00:12.530848565  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.697053611  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.697147818  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 774 : (0.144033)
0:00:12.697295530  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:12.862007878  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:12.862104134  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 774 : (0.157707)
0:00:12.862252645  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:200:gst_inceptionv4_preprocess:<net> Preprocess
0:00:13.027090881  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:232:gst_inceptionv4_postprocess:<net> Postprocess
0:00:13.027190273  4675      0x10ee590 LOG              inceptionv4 gstinceptionv4.c:253:gst_inceptionv4_postprocess:<net> Highest probability is label 774 : (0.142998)

Inceptionv4 visualization with classification overlay TensorFlow-Lite

  • Get the graph used on this example from this link
  • You will need a camera compatible with NVIDIA Libargus API or V4l2.

NVIDIA Camera

  • Pipeline
SENSOR_ID=0
MODEL_LOCATION='graph_inceptionv4.tflite'
LABELS='labels.txt'
gst-launch-1.0 \
nvcamerasrc sensor-id=$SENSOR_ID ! 'video/x-raw(memory:NVMM)' ! tee name=t \
t. ! queue max-size-buffers=1 leaky=downstream ! nvvidconv ! 'video/x-raw,format=(string)RGBA' ! net.sink_model \
t. ! queue max-size-buffers=1 leaky=downstream ! nvvidconv ! 'video/x-raw,format=(string)RGBA' ! net.sink_bypass \
inceptionv4 name=net backend=tflite model-location=$MODEL_LOCATION labels="(cat $LABELS)" \
net.src_bypass ! classificationoverlay labels="$(cat $LABELS)" font-scale=4 thickness=4 ! nvvidconv ! nvoverlaysink sync=false -v

V4L2

  • Pipeline
CAMERA='/dev/video1'
MODEL_LOCATION='graph_inceptionv4.tflite'
LABELS='labels.txt'
gst-launch-1.0 \
v4l2src device=$CAMERA ! "video/x-raw, width=1280, height=720" ! tee name=t \
t. ! videoconvert ! videoscale ! queue ! net.sink_model \
t. ! queue ! net.sink_bypass \
inceptionv4 name=net backend=tflite model-location=$MODEL_LOCATION labels="(cat $LABELS)" \
net.src_bypass ! videoconvert ! classificationoverlay labels="$(cat $LABELS)" font-scale=4 thickness=4 ! videoconvert ! xvimagesink sync=false
  • Output
Example classification overlay output

TinyYolov2 inference on image file using TensorFlow-Lite

  • Get the graph used on this example from this link
  • You will need an image file from one of TinyYOLO classes
  • Pipeline
IMAGE_FILE='cat.jpg'
MODEL_LOCATION='graph_tinyyolov2.tflite'
LABELS='labels.txt'
GST_DEBUG=tinyyolov2:6 gst-launch-1.0 \
multifilesrc location=$IMAGE_FILE ! jpegparse ! nvjpegdec ! 'video/x-raw' ! nvvidconv ! 'video/x-raw(memory:NVMM),format=NV12' ! nvvidconv ! queue ! net.sink_model \
tinyyolov2 name=net backend=tflite model-location=$MODEL_LOCATION labels="(cat $LABELS)"
  • Output
0:00:07.137677204 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.266928985 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.267080761 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:25.820670, y:11.977936, width:425.495203, height:450.224357, prob:15.204609]
0:00:07.267382968 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.394225925 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.394431653 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:25.820670, y:11.977936, width:425.495203, height:450.224357, prob:15.204609]
0:00:07.394858915 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.527547133 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.527753020 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:25.820670, y:11.977936, width:425.495203, height:450.224357, prob:15.204609]
0:00:07.528080219 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.662473455 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.662769998 30513       0x5accf0 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:25.820670, y:11.977936, width:425.495203, height:450.224357, prob:15.204609]

TinyYolov2 inference on video file using TensorFlow-Lite

  • Get the graph used on this example from this link
  • You will need a video file from one of TinyYOLO classes
  • Pipeline
VIDEO_FILE='cat.mp4'
MODEL_LOCATION='graph_tinyyolov2.tflite'
LABELS='labels.txt'
GST_DEBUG=tinyyolov2:6 gst-launch-1.0 \
filesrc location=$VIDEO_FILE ! qtdemux name=demux ! h264parse ! omxh264dec ! nvvidconv ! queue ! net.sink_model \
tinyyolov2 name=net backend=tflite model-location=$MODEL_LOCATION labels="(cat $LABELS)"
  • Output
0:00:07.245722660 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.360377432 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.360586455 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:-46.105452, y:-9.139365, width:445.139551, height:487.967720, prob:14.592537]
0:00:07.360859318 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.489190714 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.489382873 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:-46.140270, y:-9.193503, width:445.228762, height:488.028163, prob:14.596972]
0:00:07.489736216 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.629190069 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.629379733 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:-46.281640, y:-9.164348, width:445.512899, height:487.908826, prob:14.596945]
0:00:07.629717876 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:07.761072493 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:07.761271244 30545       0x5ad000 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:7, x:-46.338202, y:-9.202273, width:445.624841, height:487.954952, prob:14.592540]

TinyYolov2 inference on camera stream using TensorFlow-Lite

  • Get the graph used on this example from this link
  • You will need a camera compatible with NVIDIA Libargus API or V4l2.

NVIDIA Camera

  • Pipeline
SENSOR_ID=0
MODEL_LOCATION='graph_tinyyolov2.tflite'
LABELS='labels.txt'
GST_DEBUG=tinyyolov2:6 gst-launch-1.0 \
nvarguscamerasrc sensor-id=$SENSOR_ID ! nvvidconv ! 'video/x-raw,format=BGRx' ! queue ! net.sink_model \
tinyyolov2 name=net backend=tflite model-location=$MODEL_LOCATION labels="(cat $LABELS)"

V4L2

  • Pipeline
CAMERA='/dev/video1'
MODEL_LOCATION='graph_tinyyolov2.tflite'
LABELS='labels.txt'
GST_DEBUG=tinyyolov2:6 gst-launch-1.0 \
v4l2src device=$CAMERA ! videoconvert ! videoscale ! queue ! net.sink_model \
tinyyolov2 name=net model-location=$MODEL_LOCATION backend=tflite labels="$(cat $LABELS)"
  • Output
0:00:39.754924355  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:39.876816786  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:39.876914225  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:4, x:147.260736, y:116.184709, width:134.389472, height:245.113627, prob:8.375733]
0:00:39.877085489  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:39.999699614  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:39.999799198  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:4, x:146.957935, y:117.902112, width:134.883825, height:242.143126, prob:7.982772]
0:00:39.999962206  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:40.118613969  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:40.118712017  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:4, x:147.147349, y:116.562615, width:134.469630, height:244.181931, prob:8.139100]
0:00:40.118882641  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:479:gst_tinyyolov2_preprocess:<net> Preprocess
0:00:40.264861052  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:501:gst_tinyyolov2_postprocess:<net> Postprocess
0:00:40.264964828  5030      0x10ee590 LOG               tinyyolov2 gsttinyyolov2.c:384:print_top_predictions:<net> Box: [class:4, x:146.618516, y:117.162739, width:135.454029, height:243.785573, prob:8.112847]

TinyYolov2 visualization with detection overlay TensorFlow-Lite

  • Get the graph used on this example from this link
  • You will need a camera compatible with NVIDIA Libargus API or V4l2.

NVIDIA Camera

  • Pipeline
SENSOR_ID=0
MODEL_LOCATION='graph_tinyyolov2.tflite'
LABELS='labels.txt'
GST_DEBUG=tinyyolov2:6 \
gst-launch-1.0 \
nvcamerasrc sensor-id=$SENSOR_ID ! 'video/x-raw(memory:NVMM)' ! tee name=t \
t. ! queue max-size-buffers=1 leaky=downstream ! nvvidconv ! 'video/x-raw,format=(string)RGBA' ! net.sink_model \
t. ! queue max-size-buffers=1 leaky=downstream ! nvvidconv ! 'video/x-raw,format=(string)RGBA' ! net.sink_bypass \
tinyyolov2 name=net backend=tflite model-location=$MODEL_LOCATION labels="(cat $LABELS)" \
net.src_bypass !  detectionoverlay labels="$(cat $LABELS)" font-scale=4 thickness=4  ! nvvidconv ! nvoverlaysink sync=false -v

V4L2

  • Pipeline
CAMERA='/dev/video1'
MODEL_LOCATION='graph_tinyyolov2.tflite'
LABELS='labels.txt'
gst-launch-1.0 \
v4l2src device=$CAMERA ! "video/x-raw, width=1280, height=720" ! tee name=t \
t. ! videoconvert ! videoscale ! queue ! net.sink_model \
t. ! queue ! net.sink_bypass \
tinyyolov2 name=net backend=tflite model-location=$MODEL_LOCATION labels="(cat $LABELS)" \
net.src_bypass ! videoconvert ! detectionoverlay labels="$(cat $LABELS)" font-scale=1 thickness=2 ! videoconvert ! xvimagesink sync=false


  • Output
Example detection overlay output


Previous: Example pipelines/NANO Index Next: Example pipelines/Xavier