Start running inference right away - the following commands use pre-installed test images and require no additional setup.
Image Classification using ResNet50 INT8 model (with ONNX Runtime, internal preprocessing)
gst-launch-1.0 \
filesrc location=/etc/vai/models/resnet50_int8/data/classification.jpg ! queue \
! jpegdec \
! videoscale \
! video/x-raw,width=640,height=640 \
! videoconvert \
! video/x-raw,format=NV12 ! queue \
! vvas_xinfer config-file=/etc/vvas/configs/infer/resnet50_int8_onnxrt_intppe.json ! queue \
! vvas_xmetaconvert config-location=/etc/vvas/configs/metaconvert/metaconvert_config.json \
! vvas_xoverlay ! queue \
! filesink location=classification_out_resnet50_int8.nv12 -v
Expected output (640×640 NV12 frame with classification label overlaid):
Object Detection using YOLOXM INT8 model (with VART zero-copy, internal preprocessing)
gst-launch-1.0 \
filesrc location=/etc/vai/models/yolox_m_int8/data/detections.jpg ! queue \
! jpegdec \
! videoscale \
! video/x-raw,width=640,height=640 \
! videoconvert \
! video/x-raw,format=NV12 ! queue \
! vvas_xinfer config-file=/etc/vvas/configs/infer/yoloxm_int8_vart_zerocopy.json ! queue \
! vvas_xmetaconvert config-location=/etc/vvas/configs/metaconvert/metaconvert_config.json \
! vvas_xoverlay ! queue \
! filesink location=detection_out_yoloxm_int8.nv12 -v
Fine-tune detection behavior - confidence threshold, IoU, NMS, and more - via the postprocess-config block in the vvas_xinfer JSON. See Sample Postprocess Library Configuration Reference for the full parameter reference.