gvaroiscale#
Finds matching object-detection ROIs in GstAnalyticsRelationMeta, crops each ROI from an
RGB frame, preserves aspect ratio, and centers each resized image on a black canvas. Each
input frame produces one output buffer per selected ROI; frames without a valid ROI are dropped.
ROIs are processed in descending confidence order. Same-class detections with an
intersection-over-union (IoU) above 0.5 are deduplicated, keeping the detection with the
highest confidence.
If an input buffer carries the marked shared ROI created by gvaframeselect, that ROI takes
precedence over the detector boxes and produces one crop. In that case, target-class and
confidence filtering has already been applied by gvaframeselect.
Property |
Type |
Range |
Default |
Description |
|---|---|---|---|---|
|
String |
— |
(empty) |
Comma-separated detection labels to consider; an empty value allows all classes. |
|
Double |
[0.0, 1.0] |
0.0 |
Minimum detection confidence required to select an ROI. |
|
Unsigned Integer |
[1, 8192] |
448 |
Output width; configure for the model input expected by the next element. |
|
Unsigned Integer |
[1, 8192] |
448 |
Output height; configure for the model input expected by the next element. |
|
String |
— |
(empty) |
If set, each scaled ROI frame is also saved as |
The element accepts and outputs video/x-raw,format=RGB. Output dimensions are negotiated
from out-width and out-height.
Set these dimensions according to the downstream model’s input requirements. Models that accept
dynamic image sizes or resize inputs internally may not require an exact match; check the model
documentation. The defaults are not universal.
The output directory is optional; an empty value disables image saving. File numbering advances
for each crop, starts at zero, and restarts on each new stream. Every output crop carries a
GstAnalyticsODMtd with the class and confidence of its source detection; its box is the area the
scaled ROI occupies inside the output frame, excluding the black padding. Outputs from the same input
frame retain its PTS, duration and buffer offset.
When target-classes is set, only ROIs whose labels match one of the comma-separated values are
considered. Whitespace around labels is ignored.
... ! gvaframeselect frame-budget=8 target-classes=person,vehicle confidence-threshold=0.5 ! \
videoconvert ! video/x-raw,format=RGB ! \
gvaroiscale out-width=448 out-height=448 target-classes=person,vehicle ! gvagenai ...
For a runnable command-line example, see the gvaroiscale sample.
Element Details (gst-inspect-1.0)#
Pad Templates:
SINK template: 'sink'
Availability: Always
Capabilities:
video/x-raw
format: RGB (gchararray)
width: [ 1, 8192 ] (GstIntRange)
height: [ 1, 8192 ] (GstIntRange)
SRC template: 'src'
Availability: Always
Capabilities:
video/x-raw
format: RGB (gchararray)
width: [ 1, 8192 ] (GstIntRange)
height: [ 1, 8192 ] (GstIntRange)
Element has no clocking capabilities.
Element has no URI handling capabilities.
Pads:
SINK: 'sink'
Pad Template: 'sink'
SRC: 'src'
Pad Template: 'src'
Element Properties:
confidence-threshold: Minimum detection confidence required to select an ROI
flags: readable, writable
Double. Range: 0 - 1 Default: 0
name : The name of the object
flags: readable, writable
String. Default: "gvaroiscale0"
out-height : Scaled ROI output height
flags: readable, writable
Unsigned Integer. Range: 1 - 8192 Default: 448
out-width : Scaled ROI output width
flags: readable, writable
Unsigned Integer. Range: 1 - 8192 Default: 448
output-dir : If set, each scaled ROI frame is also saved as a JPEG in this directory
flags: readable, writable
String. Default: null
parent : The parent of the object
flags: readable, writable
Object of type "GstObject"
qos : Handle Quality-of-Service events
flags: readable, writable
Boolean. Default: false
target-classes : Comma-separated detection labels to consider; empty means all classes
flags: readable, writable
String. Default: ""