burn-yoloxinference

burn-based inference element that performs YOLOX object detection.

 gst-launch-1.0 souphttpsrc location=https://raw.githubusercontent.com/tracel-ai/models/ab8c64bd7e1f45e99cc321ce900a5b5e6b97910c/yolox-burn/samples/dog_bike_man.jpg \
     ! jpegdec ! videoconvertscale ! "video/x-raw,width=800,height=640" \
     ! burn-yoloxinference ! yoloxtensordec label-file=COCO_classes.txt \
     ! videoconvertscale ! objectdetectionoverlay \
     ! videoconvertscale ! imagefreeze ! autovideosink -v
 |] This takes a JPEG, performs object detection via `burn-yoloxinference` on it, decodes the
 inferred tensors with `yoloxtensordec` and then overlays the detected objects on the frame via
 `objectdetectionoverlay`.

Hierarchy

GObject
    ╰──GInitiallyUnowned
        ╰──GstObject
            ╰──GstElement
                ╰──GstBaseTransform
                    ╰──burn-yoloxinference

Factory details

Authors: – Sebastian Dröge

Classification: – Inference/Classification/Video

Rank – none

Plugin – burn

Package – gst-plugin-burn

Pad Templates

sink

video/x-raw:
         format: RGB
          width: [ 32, 2147483616, 32 ]
         height: [ 32, 2147483616, 32 ]
      framerate: [ 0/1, 2147483647/1 ]
pixel-aspect-ratio: 1/1

Presence – always

Direction – sink

Object type – GstPad

src

video/x-raw:
         format: RGB
          width: [ 32, 2147483616, 32 ]
         height: [ 32, 2147483616, 32 ]
      framerate: [ 0/1, 2147483647/1 ]
pixel-aspect-ratio: 1/1
        tensors: "tensorgroups\,\ yolox-out\=\(/uniquelist\)\{\ \(caps\)\"tensor/strided\\\,\\\ dims\\\=\\\(int\\\)\\\<\\\ 1\\\,\\\ 0\\\,\\\ \\\[\\\ 5\\\,\\\ 2147483647\\\ \\\]\\\ \\\>\\\,\\\ dims-order\\\=\\\(string\\\)row-major\\\,\\\ type\\\=\\\(string\\\)float32\"\ \}\;"

Presence – always

Direction – src

Object type – GstPad

Properties

backend-type 

“backend-type” GstBurnBackendType *

Burn backend to use

Flags : Read / Write

Default value : nd-array (0)

cubecl-index-id 

“cubecl-index-id” guint

Index ID that identifies the device number. For CubeCL-based backends only.

Flags : Read / Write

Default value : -1

cubecl-type-id 

“cubecl-type-id” guint

Type ID that identifies the type of the device. For CubeCL-based backends only, -1 for default.

Flags : Read / Write

Default value : -1

model-type 

“model-type” GstBurnYoloxModelType *

YOLOX model type to use

Flags : Read / Write

Default value : tiny (1)

num-classes 

“num-classes” guint

Number of output classes of the model. This must match the weights. Keep at 0 for pretrained models.

Flags : Read / Write

Default value : 0

weights-path 

“weights-path” gchararray

Path to a PyTorch weights file for the model. This must match the model type and number of weights. Keep empty for pretrained models.

Flags : Read / Write

Default value : NULL

Named constants

GstBurnBackendType

Backend that should be used. The NdArray backend is always available and is a CPU-backed backend.

Available backends depend on build-time options.

Members

nd-array (0) – NdArray

Since : plugins-rs-0.15.0

GstBurnYoloxModelType

YOLOX model that should be used.

Members

nano (0) – Nano
tiny (1) – Tiny
small (2) – Small
medium (3) – Medium
large (4) – Large
extra-large (5) – ExtraLarge

Since : plugins-rs-0.15.0

The results of the search are