Udabur Investment：C# onnx Detic detects 20,000 types of objects

github address:

Code Release for "Detecting Twenty-Thousand Classes USIGE-Level Supervision".

Features

Detects Any Class Given Class Names (USING CLIP).

We Train the Detector on ImageNet-21k DataSet with 21k Classes.

Cross-Dataset Generalization to Openimages and Objects365 Without Finetuning.

State-the-The-Vocabular Lvis and Open-Vocabular Coco.Udabur Investment

Works for Detr-Style Detectors.

Inputs

---------------------------

name: IMG

tensor: Float [1, 3, -1, -1]

----------------------------------------------------------------

Outputs

---------------------------

name: Pred_boxes

tensor: Float [-1, 4]

name: scores

tensor: float [-1]

name: Pred_Classes

tensor: int64 [-1]

name: Pred_masks

tensor: Float [-1, 1, -1, -1]

----------------------------------------------------------------

Vs2022

.NET Framework 4.8

Opencvsharp 4.8

Microsoft.ml.onnxruntime 1.16.2

var pred_boxes = results_onnxvalue [0] .stensor (). Toarray ();

var scores = results_onnxValue [1] .stensor (). Toarray ();

var pred_classes = results_onnxvalue [2] .stensor ()Jaipur Wealth Management. Toarray ();

varter_masks = results_onnxValue [3] .stensor (). Toarray ();

int Num_box = results_onnxValue [0] .astensor (). Dimensions [0];

float scale_x = factors [0];

float scale_y = factors [1];

List preds = new list ();

For (int i = 0; I threshold && health> Threshold)

Preds.add (New Boxinfo (XMIN, Ymin, XMAX, YMAX, Score, Class_name));