gstanalytics : Add tensor decoder element for yolo detection models

Existing tensor decoder has been bifurcated into two seperate gst
element as:

`yoloodv5tensordecoder`: decodes tensors output(masks) from detection-only
models e.g yolov8s.onnx

`yolsegv8tensordecoder`: decoder tensors output(masks and logits) from
segementation models e.g FastSAM or yolov8s-seg
This commit is contained in:
Santosh Mahto 2025-03-11 20:50:05 +05:30 committed by Elias Rosendahl
parent 87b56fbf86
commit 62731c958c
3 changed files with 726 additions and 485 deletions

View File

@ -39,7 +39,8 @@ plugin_init (GstPlugin * plugin)
{
gboolean ret = FALSE;
ret |= GST_ELEMENT_REGISTER (ssd_object_detector, plugin);
ret |= GST_ELEMENT_REGISTER (yolo_tensor_decoder, plugin);
ret |= GST_ELEMENT_REGISTER (yolo_seg_tensor_decoder, plugin);
ret |= GST_ELEMENT_REGISTER (yolo_od_tensor_decoder, plugin);
return ret;
}

View File

@ -33,9 +33,20 @@
G_BEGIN_DECLS
#define GST_TYPE_YOLO_TENSOR_DECODER (gst_yolo_tensor_decoder_get_type ())
G_DECLARE_FINAL_TYPE (GstYoloTensorDecoder, gst_yolo_tensor_decoder,
GST, YOLO_TENSOR_DECODER, GstBaseTransform)
#define GST_TYPE_YOLO_OD_TENSOR_DECODER (gst_yolo_od_tensor_decoder_get_type ())
#define GST_YOLO_OD_TENSOR_DECODER(obj) \
(G_TYPE_CHECK_INSTANCE_CAST ((obj), GST_TYPE_YOLO_OD_TENSOR_DECODER, GstYoloOdTensorDecoder))
#define GST_YOLO_OD_TENSOR_DECODER_CLASS(klass) \
(G_TYPE_CHECK_CLASS_CAST ((klass), GST_TYPE_YOLO_OD_TENSOR_DECODER, GstYoloOdTensorDecoderClass))
#define GST_IS_YOLO_OD_TENSOR_DECODER(obj) \
(G_TYPE_CHECK_INSTANCE_TYPE ((obj), GST_TYPE_YOLO_OD_TENSOR_DECODER))
#define GST_IS_YOLO_OD_TENSOR_DECODER_CLASS(klass) \
(G_TYPE_CHECK_CLASS_TYPE ((klass), GST_TYPE_YOLO_OD_TENSOR_DECODER))
#define GST_YOLO_OD_TENSOR_DECODER_GET_CLASS(obj) \
(G_TYPE_INSTANCE_GET_CLASS ((obj), GST_TYPE_YOLO_OD_TENSOR_DECODER, GstYoloOdTensorDecoderClass))
typedef struct _GstYoloOdTensorDecoder GstYoloOdTensorDecoder;
typedef struct _GstYoloOdTensorDecoderClass GstYoloOdTensorDecoderClass;
typedef struct _BBox
{
@ -45,7 +56,7 @@ typedef struct _BBox
guint h;
} BBox;
struct _GstYoloTensorDecoder
struct _GstYoloOdTensorDecoder
{
GstBaseTransform basetransform;
/* Box confidence threshold */
@ -57,52 +68,60 @@ struct _GstYoloTensorDecoder
/* Maximum detection/mask */
gsize max_detection;
/* Video Info */
GstVideoInfo video_info;
/* Candidates with a class confidence level above threshold. */
GPtrArray *sel_candidates;
/* Final candidates selected that respect class confidence level,
* NMS and maximum detection. */
GPtrArray *selected;
/* Tensor-id identifying mask tensors out of yolo inference process. */
GQuark mask_tensor_id;
/* Tensor-id identifying logits tensors out of yolo inference process. */
GQuark logits_tensor_id;
/* Region of the mask that contain valid segmentation information */
BBox mask_roi;
/* Scaling factor to convert bounding-box coordinates to mask coordinates */
gfloat bb2mask_gain;
/* Mask width */
guint mask_w;
/* Mask height */
guint mask_h;
/* Mask length */
gsize mask_length;
/* BufferPool for mask */
GstBufferPool *mask_pool;
GstVideoInfo video_info;
/* Labels file */
gchar *label_file;
/* Labels */
GArray *labels;
/* GstAnalyticsODMtd data */
GArray *od_mtds;
/* Hash table to store the offset in the mask tensor buffer where
* OdMtd data are stored. key is OdMtd.id */
GHashTable *candidate_offsets;
};
struct _GstYoloTensorDecoderClass
struct _GstYoloOdTensorDecoderClass
{
GstBaseTransformClass parent_class;
};
GST_ELEMENT_REGISTER_DECLARE (yolo_tensor_decoder)
GType gst_yolo_od_tensor_decoder_get_type (void);
G_DEFINE_AUTOPTR_CLEANUP_FUNC (GstYoloOdTensorDecoder, g_object_unref)
GST_ELEMENT_REGISTER_DECLARE (yolo_od_tensor_decoder)
/* Yolo segmentation tensor decoder */
#define GST_TYPE_YOLO_SEG_TENSOR_DECODER (gst_yolo_seg_tensor_decoder_get_type ())
G_DECLARE_FINAL_TYPE (GstYoloSegTensorDecoder, gst_yolo_seg_tensor_decoder,
GST, YOLO_SEG_TENSOR_DECODER, GstYoloOdTensorDecoder)
struct _GstYoloSegTensorDecoder
{
GstYoloOdTensorDecoder parent;
/* Mask width */
guint mask_w;
/* Mask height */
guint mask_h;
/* Mask length */
gsize mask_length;
GQuark logits_tensor_id;
gfloat bb2mask_gain;
/* BufferPool for mask */
BBox mask_roi;
/* BufferPool for mask */
GstBufferPool *mask_pool;
};
GST_ELEMENT_REGISTER_DECLARE (yolo_seg_tensor_decoder)
G_END_DECLS
#endif /* __GST_YOLO_TENSOR_DECODER_H__ */