private static int DetectSqueezeNet(NcnnDotNet.OpenCV.Mat bgr, List <float> clsScores)
{
using (var squeezeNet = new Net())
{
if (Ncnn.IsSupportVulkan)
{
squeezeNet.Opt.UseVulkanCompute = true;
}
// the ncnn model https://github.com/nihui/ncnn-assets/tree/master/models
squeezeNet.LoadParam("squeezenet_v1.1.param");
squeezeNet.LoadModel("squeezenet_v1.1.bin");
using var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr, bgr.Cols, bgr.Rows, 227, 227);
var meanVals = new[] { 104f, 117f, 123f };
@in.SubstractMeanNormalize(meanVals, null);
using var ex = squeezeNet.CreateExtractor();
ex.Input("data", @in);
using var @out = new Mat();
ex.Extract("prob", @out);
clsScores.Capacity = @out.W;
for (var j = 0; j < @out.W; j++)
{
clsScores.Add(@out[j]);
}
}
return(0);
}
private static int DetectYoloV3(NcnnDotNet.OpenCV.Mat bgr, List <Object> objects)
{
using (var yolov3 = new Net())
{
if (Ncnn.IsSupportVulkan)
{
yolov3.Opt.UseVulkanCompute = true;
}
// original pretrained model from https://github.com/eric612/MobileNet-YOLO
// param : https://drive.google.com/open?id=1V9oKHP6G6XvXZqhZbzNKL6FI_clRWdC-
// bin : https://drive.google.com/open?id=1DBcuFCr-856z3FRQznWL_S5h-Aj3RawA
// the ncnn model https://github.com/nihui/ncnn-assets/tree/master/models
yolov3.LoadParam("mobilenetv2_yolov3.param");
yolov3.LoadModel("mobilenetv2_yolov3.bin");
const int targetSize = 352;
var imgW = bgr.Cols;
var imgH = bgr.Rows;
using (var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr, bgr.Cols, bgr.Rows, targetSize, targetSize))
{
var meanVals = new[] { 127.5f, 127.5f, 127.5f };
var normVals = new[] { 0.007843f, 0.007843f, 0.007843f };
@in.SubstractMeanNormalize(meanVals, normVals);
using (var ex = yolov3.CreateExtractor())
{
ex.SetNumThreads(4);
ex.Input("data", @in);
using (var @out = new Mat())
{
ex.Extract("detection_out", @out);
// printf("%d %d %d\n", out.w, out.h, out.c);
objects.Clear();
for (var i = 0; i < @out.H; i++)
{
var values = @out.Row(i);
var @object = new Object();
@object.Label = (int)values[0];
@object.Prob = values[1];
@object.Rect.X = values[2] * imgW;
@object.Rect.Y = values[3] * imgH;
@object.Rect.Width = values[4] * imgW - @object.Rect.X;
@object.Rect.Height = values[5] * imgH - @object.Rect.Y;
objects.Add(@object);
}
}
}
}
}
return(0);
}
private static int DetectMobileNetV3(NcnnDotNet.OpenCV.Mat bgr, List <Object> objects)
{
using (var mobilenetV3 = new Net())
{
if (Ncnn.IsSupportVulkan)
{
mobilenetV3.Opt.UseVulkanCompute = true;
}
// converted ncnn model from https://github.com/ujsyehao/mobilenetv3-ssd
mobilenetV3.LoadParam("mobilenetv3_ssdlite_voc.param");
mobilenetV3.LoadModel("mobilenetv3_ssdlite_voc.bin");
const int targetSize = 300;
var imgW = bgr.Cols;
var imgH = bgr.Rows;
using var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr2Rgb, bgr.Cols, bgr.Rows, targetSize, targetSize);
var meanVals = new[] { 123.675f, 116.28f, 103.53f };
var normVals = new[] { 1.0f, 1.0f, 1.0f };
@in.SubstractMeanNormalize(meanVals, normVals);
using var ex = mobilenetV3.CreateExtractor();
ex.SetLiteMode(true);
ex.SetNumThreads(4);
ex.Input("input", @in);
using var @out = new Mat();
ex.Extract("detection_out", @out);
// printf("%d %d %d\n", out.w, out.h, out.c);
objects.Clear();
for (var i = 0; i < @out.H; i++)
{
var values = @out.Row(i);
var @object = new Object();
@object.Label = (int)values[0];
@object.Prob = values[1];
// filter out cross-boundary
var x1 = Clamp(values[2] * targetSize, 0.0f, targetSize - 1) / targetSize * imgW;
var y1 = Clamp(values[3] * targetSize, 0.0f, targetSize - 1) / targetSize * imgH;
var x2 = Clamp(values[4] * targetSize, 0.0f, targetSize - 1) / targetSize * imgW;
var y2 = Clamp(values[5] * targetSize, 0.0f, targetSize - 1) / targetSize * imgH;
@object.Rect.X = x1;
@object.Rect.Y = y1;
@object.Rect.Width = x2 - x1;
@object.Rect.Height = y2 - y1;
objects.Add(@object);
}
}
return(0);
}
private static int DetectYoloV2(NcnnDotNet.OpenCV.Mat bgr, List <Object> objects)
{
using (var yolov2 = new Net())
{
if (Ncnn.IsSupportVulkan)
{
yolov2.Opt.UseVulkanCompute = true;
}
// original pretrained model from https://github.com/eric612/MobileNet-YOLO
// https://github.com/eric612/MobileNet-YOLO/blob/master/models/yolov2/mobilenet_yolo_deploy.prototxt
// https://github.com/eric612/MobileNet-YOLO/blob/master/models/yolov2/mobilenet_yolo_deploy_iter_80000.caffemodel
// the ncnn model https://github.com/nihui/ncnn-assets/tree/master/models
yolov2.LoadParam("mobilenet_yolo.param");
yolov2.LoadModel("mobilenet_yolo.bin");
const int targetSize = 416;
var imgW = bgr.Cols;
var imgH = bgr.Rows;
using var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr, bgr.Cols, bgr.Rows, targetSize, targetSize);
// the Caffe-YOLOv2-Windows style
// X' = X * scale - mean
var meanVals = new[] { 1.0f, 1.0f, 1.0f };
var normVals = new[] { 0.007843f, 0.007843f, 0.007843f };
@in.SubstractMeanNormalize(null, normVals);
@in.SubstractMeanNormalize(meanVals, null);
using var ex = yolov2.CreateExtractor();
ex.SetNumThreads(4);
ex.Input("data", @in);
using var @out = new Mat();
ex.Extract("detection_out", @out);
// printf("%d %d %d\n", out.w, out.h, out.c);
objects.Clear();
for (var i = 0; i < @out.H; i++)
{
var values = @out.Row(i);
var @object = new Object();
@object.Label = (int)values[0];
@object.Prob = values[1];
@object.Rect.X = values[2] * imgW;
@object.Rect.Y = values[3] * imgH;
@object.Rect.Width = values[4] * imgW - @object.Rect.X;
@object.Rect.Height = values[5] * imgH - @object.Rect.Y;
objects.Add(@object);
}
}
return(0);
}
private static int DetectPeleeNet(NcnnDotNet.OpenCV.Mat bgr, List <Object> objects, NcnnDotNet.Mat resized)
{
using (var peleenet = new Net())
{
if (Ncnn.IsSupportVulkan)
{
peleenet.Opt.UseVulkanCompute = true;
}
// model is converted from https://github.com/eric612/MobileNet-YOLO
// and can be downloaded from https://drive.google.com/open?id=1Wt6jKv13sBRMHgrGAJYlOlRF-o80pC0g
// the ncnn model https://github.com/nihui/ncnn-assets/tree/master/models
peleenet.LoadParam("pelee.param");
peleenet.LoadModel("pelee.bin");
const int targetSize = 300;
var imgW = bgr.Cols;
var imgH = bgr.Rows;
using var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr, bgr.Cols, bgr.Rows, targetSize, targetSize);
var meanVals = new[] { 103.9f, 116.7f, 123.6f };
var normVals = new[] { 0.017f, 0.017f, 0.017f };
@in.SubstractMeanNormalize(meanVals, normVals);
using var ex = peleenet.CreateExtractor();
//ex.SetNumThreads(4);
ex.Input("data", @in);
using var @out = new Mat();
ex.Extract("detection_out", @out);
// printf("%d %d %d\n", out.w, out.h, out.c);
objects.Clear();
for (var i = 0; i < @out.H; i++)
{
var values = @out.Row(i);
var @object = new Object();
@object.Label = (int)values[0];
@object.Prob = values[1];
@object.Rect.X = values[2] * imgW;
@object.Rect.Y = values[3] * imgH;
@object.Rect.Width = values[4] * imgW - @object.Rect.X;
@object.Rect.Height = values[5] * imgH - @object.Rect.Y;
objects.Add(@object);
}
using var segOut = new Mat();
ex.Extract("sigmoid", segOut);
Ncnn.ResizeBilinear(segOut, resized, imgW, imgH);
}
return(0);
}
private static int DetectMobileNetV2(NcnnDotNet.OpenCV.Mat bgr, List <Object> objects)
{
using (var mobilenetV2 = new Net())
{
if (Ncnn.IsSupportVulkan)
{
mobilenetV2.Opt.UseVulkanCompute = true;
}
// ToDo: Support Custom Layer from C#
//mobilenetV2.register_custom_layer("Silence", Noop_layer_creator);
// original pretrained model from https://github.com/chuanqi305/MobileNetv2-SSDLite
// https://github.com/chuanqi305/MobileNetv2-SSDLite/blob/master/ssdlite/voc/deploy.prototxt
// the ncnn model https://github.com/nihui/ncnn-assets/tree/master/models
mobilenetV2.LoadParam("mobilenetv2_ssdlite_voc.param");
mobilenetV2.LoadModel("mobilenetv2_ssdlite_voc.bin");
const int targetSize = 300;
var imgW = bgr.Cols;
var imgH = bgr.Rows;
using var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr, bgr.Cols, bgr.Rows, targetSize, targetSize);
var meanVals = new[] { 127.5f, 127.5f, 127.5f };
var normVals = new[] { (float)(1.0 / 127.5), (float)(1.0 / 127.5), (float)(1.0 / 127.5) };
@in.SubstractMeanNormalize(meanVals, normVals);
using var ex = mobilenetV2.CreateExtractor();
ex.SetLiteMode(true);
ex.SetNumThreads(4);
ex.Input("data", @in);
using var @out = new Mat();
ex.Extract("detection_out", @out);
// printf("%d %d %d\n", out.w, out.h, out.c);
objects.Clear();
for (var i = 0; i < @out.H; i++)
{
var values = @out.Row(i);
var @object = new Object();
@object.Label = (int)values[0];
@object.Prob = values[1];
@object.Rect.X = values[2] * imgW;
@object.Rect.Y = values[3] * imgH;
@object.Rect.Width = values[4] * imgW - @object.Rect.X;
@object.Rect.Height = values[5] * imgH - @object.Rect.Y;
objects.Add(@object);
}
}
return(0);
}
private static int DetectSqueezeNet(NcnnDotNet.OpenCV.Mat bgr, List <Object> objects)
{
using (var squeezenet = new Net())
{
if (Ncnn.IsSupportVulkan)
{
squeezenet.Opt.UseVulkanCompute = true;
}
// original pretrained model from https://github.com/chuanqi305/SqueezeNet-SSD
// squeezenet_ssd_voc_deploy.prototxt
// https://drive.google.com/open?id=0B3gersZ2cHIxdGpyZlZnbEQ5Snc
// the ncnn model https://github.com/nihui/ncnn-assets/tree/master/models
squeezenet.LoadParam("squeezenet_ssd_voc.param");
squeezenet.LoadModel("squeezenet_ssd_voc.bin");
const int targetSize = 300;
var imgW = bgr.Cols;
var imgH = bgr.Rows;
using var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr, bgr.Cols, bgr.Rows, targetSize, targetSize);
var meanVals = new[] { 104f, 117f, 123f };
@in.SubstractMeanNormalize(meanVals, null);
using var ex = squeezenet.CreateExtractor();
ex.SetNumThreads(4);
ex.Input("data", @in);
using var @out = new Mat();
ex.Extract("detection_out", @out);
// printf("%d %d %d\n", out.w, out.h, out.c);
objects.Clear();
for (var i = 0; i < @out.H; i++)
{
var values = @out.Row(i);
var @object = new Object();
@object.Label = (int)values[0];
@object.Prob = values[1];
@object.Rect.X = values[2] * imgW;
@object.Rect.Y = values[3] * imgH;
@object.Rect.Width = values[4] * imgW - @object.Rect.X;
@object.Rect.Height = values[5] * imgH - @object.Rect.Y;
objects.Add(@object);
}
}
return(0);
}
private static int DetectMobileNet(NcnnDotNet.OpenCV.Mat bgr, List <Object> objects)
{
using (var mobilenet = new Net())
{
if (Ncnn.IsSupportVulkan)
{
mobilenet.Opt.UseVulkanCompute = true;
}
// model is converted from https://github.com/chuanqi305/MobileNet-SSD
// and can be downloaded from https://drive.google.com/open?id=0ByaKLD9QaPtucWk0Y0dha1VVY0U
// the ncnn model https://github.com/nihui/ncnn-assets/tree/master/models
mobilenet.LoadParam("mobilenet_ssd_voc_ncnn.param");
mobilenet.LoadModel("mobilenet_ssd_voc_ncnn.bin");
const int targetSize = 300;
var imgW = bgr.Cols;
var imgH = bgr.Rows;
using var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr, bgr.Cols, bgr.Rows, targetSize, targetSize);
var meanVals = new[] { 127.5f, 127.5f, 127.5f };
var normVals = new[] { (float)(1.0 / 127.5), (float)(1.0 / 127.5), (float)(1.0 / 127.5) };
@in.SubstractMeanNormalize(meanVals, normVals);
using var ex = mobilenet.CreateExtractor();
//ex.SetNumThreads(4);
ex.Input("data", @in);
using var @out = new Mat();
ex.Extract("detection_out", @out);
// printf("%d %d %d\n", out.w, out.h, out.c);
objects.Clear();
for (var i = 0; i < @out.H; i++)
{
var values = @out.Row(i);
var @object = new Object();
@object.Label = (int)values[0];
@object.Prob = values[1];
@object.Rect.X = values[2] * imgW;
@object.Rect.Y = values[3] * imgH;
@object.Rect.Width = values[4] * imgW - @object.Rect.X;
@object.Rect.Height = values[5] * imgH - @object.Rect.Y;
objects.Add(@object);
}
}
return(0);
}
private static int DetectShuffleNetV2(NcnnDotNet.OpenCV.Mat bgr, List <float> clsScores)
{
using (var shuffleNetV2 = new Net())
{
if (Ncnn.IsSupportVulkan)
{
shuffleNetV2.Opt.UseVulkanCompute = true;
}
// https://github.com/miaow1988/ShuffleNet_V2_pytorch_caffe
// models can be downloaded from https://github.com/miaow1988/ShuffleNet_V2_pytorch_caffe/releases
shuffleNetV2.LoadParam("shufflenet_v2_x0.5.param");
shuffleNetV2.LoadModel("shufflenet_v2_x0.5.bin");
using var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr, bgr.Cols, bgr.Rows, 224, 224);
var normVals = new[] { 1 / 255.0f, 1 / 255.0f, 1 / 255.0f };
@in.SubstractMeanNormalize(null, normVals);
using var ex = shuffleNetV2.CreateExtractor();
ex.Input("data", @in);
using var @out = new Mat();
ex.Extract("fc", @out);
// manually call softmax on the fc output
// convert result into probability
// skip if your model already has softmax operation
{
using var softmax = Ncnn.CreateLayer("Softmax");
using var pd = new ParamDict();
softmax.LoadParam(pd);
softmax.ForwardInplace(@out, shuffleNetV2.Opt);
}
using var @out2 = @out.Reshape(@out.W * @out.H * @out.C);
clsScores.Capacity = @out2.W;
for (var j = 0; j < @out2.W; j++)
{
clsScores.Add(@out2[j]);
}
}
return(0);
}
private static int DetectRetinaFace(NcnnDotNet.OpenCV.Mat bgr, List <FaceObject> faceObjects)
{
using (var retinaFace = new Net())
{
if (Ncnn.IsSupportVulkan)
{
retinaFace.Opt.UseVulkanCompute = true;
}
// model is converted from
// https://github.com/deepinsight/insightface/tree/master/RetinaFace#retinaface-pretrained-models
// https://github.com/deepinsight/insightface/issues/669
// the ncnn model https://github.com/nihui/ncnn-assets/tree/master/models
// retinaface.load_param("retinaface-R50.param");
// retinaface.load_model("retinaface-R50.bin");
retinaFace.LoadParam("mnet.25-opt.param");
retinaFace.LoadModel("mnet.25-opt.bin");
const float probThreshold = 0.8f;
const float nmsThreshold = 0.4f;
var imgW = bgr.Cols;
var imgH = bgr.Rows;
using var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr2Rgb, bgr.Cols, bgr.Rows, imgW, imgH);
using var ex = retinaFace.CreateExtractor();
ex.Input("data", @in);
var faceProposals = new List <FaceObject>();
// stride 32
{
using var scoreBlob = new Mat();
using var bboxBlob = new Mat();
using var landmarkBlob = new Mat();
ex.Extract("face_rpn_cls_prob_reshape_stride32", scoreBlob);
ex.Extract("face_rpn_bbox_pred_stride32", bboxBlob);
ex.Extract("face_rpn_landmark_pred_stride32", landmarkBlob);
const int baseSize = 16;
const int featStride = 32;
using var ratios = new Mat(1);
ratios[0] = 1.0f;
using var scales = new Mat(2);
scales[0] = 32.0f;
scales[1] = 16.0f;
using var anchors = GenerateAnchors(baseSize, ratios, scales);
var faceObjects32 = new List <FaceObject>();
GenerateProposals(anchors, featStride, scoreBlob, bboxBlob, landmarkBlob, probThreshold, faceObjects32);
faceProposals.AddRange(faceObjects32);
}
// stride 16
{
using var scoreBlob = new Mat();
using var bboxBlob = new Mat();
using var landmarkBlob = new Mat();
ex.Extract("face_rpn_cls_prob_reshape_stride16", scoreBlob);
ex.Extract("face_rpn_bbox_pred_stride16", bboxBlob);
ex.Extract("face_rpn_landmark_pred_stride16", landmarkBlob);
const int baseSize = 16;
const int featStride = 16;
using var ratios = new Mat(1);
ratios[0] = 1.0f;
using var scales = new Mat(2);
scales[0] = 8.0f;
scales[1] = 4.0f;
using var anchors = GenerateAnchors(baseSize, ratios, scales);
var faceObjects16 = new List <FaceObject>();
GenerateProposals(anchors, featStride, scoreBlob, bboxBlob, landmarkBlob, probThreshold, faceObjects16);
faceProposals.AddRange(faceObjects16);
}
// stride 8
{
using var scoreBlob = new Mat();
using var bboxBlob = new Mat();
using var landmarkBlob = new Mat();
ex.Extract("face_rpn_cls_prob_reshape_stride8", scoreBlob);
ex.Extract("face_rpn_bbox_pred_stride8", bboxBlob);
ex.Extract("face_rpn_landmark_pred_stride8", landmarkBlob);
const int baseSize = 16;
const int featStride = 8;
using var ratios = new Mat(1);
ratios[0] = 1.0f;
using var scales = new Mat(2);
scales[0] = 2.0f;
scales[1] = 1.0f;
using var anchors = GenerateAnchors(baseSize, ratios, scales);
var faceObjects8 = new List <FaceObject>();
GenerateProposals(anchors, featStride, scoreBlob, bboxBlob, landmarkBlob, probThreshold, faceObjects8);
faceProposals.AddRange(faceObjects8);
}
// sort all proposals by score from highest to lowest
QsortDescentInplace(faceProposals);
// apply nms with nms_threshold
var picked = new List <int>();
NmsSortedBBoxes(faceProposals, picked, nmsThreshold);
var faceCount = picked.Count;
// resolve point from heatmap
faceObjects.AddRange(new FaceObject[faceCount]);
for (var i = 0; i < faceCount; i++)
{
faceObjects[i] = faceProposals[picked[i]];
// clip to image size
var x0 = faceProposals[i].Rect.X;
var y0 = faceProposals[i].Rect.Y;
var x1 = x0 + faceProposals[i].Rect.Width;
var y1 = y0 + faceProposals[i].Rect.Height;
x0 = Math.Max(Math.Min(x0, (float)imgW - 1), 0.0f);
y0 = Math.Max(Math.Min(y0, (float)imgH - 1), 0.0f);
x1 = Math.Max(Math.Min(x1, (float)imgW - 1), 0.0f);
y1 = Math.Max(Math.Min(y1, (float)imgH - 1), 0.0f);
faceObjects[i].Rect.X = x0;
faceObjects[i].Rect.Y = y0;
faceObjects[i].Rect.Width = x1 - x0;
faceObjects[i].Rect.Height = y1 - y0;
}
}
return(0);
}
private static int DetectFasterRCNN(NcnnDotNet.OpenCV.Mat bgr, List <Object> objects)
{
using (var fasterRcnn = new Net())
{
if (Ncnn.IsSupportVulkan)
{
fasterRcnn.Opt.UseVulkanCompute = true;
}
// original pretrained model from https://github.com/rbgirshick/py-faster-rcnn
// py-faster-rcnn/models/pascal_voc/ZF/faster_rcnn_alt_opt/faster_rcnn_test.pt
// https://dl.dropboxusercontent.com/s/o6ii098bu51d139/faster_rcnn_models.tgz?dl=0
// ZF_faster_rcnn_final.caffemodel
// the ncnn model https://github.com/nihui/ncnn-assets/tree/master/models
fasterRcnn.LoadParam("ZF_faster_rcnn_final.param");
fasterRcnn.LoadModel("ZF_faster_rcnn_final.bin");
// hyper parameters taken from
// py-faster-rcnn/lib/fast_rcnn/config.py
// py-faster-rcnn/lib/fast_rcnn/test.py
const int targetSize = 600; // __C.TEST.SCALES
const int maxPerImage = 100;
const float confidenceThresh = 0.05f;
const float nmsThreshold = 0.3f;// __C.TEST.NMS
// scale to target detect size
var w = bgr.Cols;
var h = bgr.Rows;
float scale;
if (w < h)
{
scale = (float)targetSize / w;
w = targetSize;
h = (int)(h * scale);
}
else
{
scale = (float)targetSize / h;
h = targetSize;
w = (int)(w * scale);
}
using var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr, bgr.Cols, bgr.Rows, w, h);
var meanVals = new[] { 102.9801f, 115.9465f, 122.7717f };
@in.SubstractMeanNormalize(meanVals, null);
using var im_info = new Mat(3);
im_info[0] = h;
im_info[1] = w;
im_info[2] = scale;
// step1, extract feature and all rois
using var ex1 = fasterRcnn.CreateExtractor();
ex1.Input("data", @in);
ex1.Input("im_info", im_info);
using var conv5Relu5 = new Mat(); // feature
using var rois = new Mat(); // all rois
ex1.Extract("conv5_relu5", conv5Relu5);
ex1.Extract("rois", rois);
// step2, extract bbox and score for each roi
var classCandidates = new List <List <Object> >();
for (var i = 0; i < rois.C; i++)
{
using var ex2 = fasterRcnn.CreateExtractor();
using var roi = rois.Channel(i); // get single roi
ex2.Input("conv5_relu5", conv5Relu5);
ex2.Input("rois", roi);
using var bboxPred = new Mat();
using var clsProb = new Mat();
ex2.Extract("bbox_pred", bboxPred);
ex2.Extract("cls_prob", clsProb);
var numClass = clsProb.W;
// There is no equivalent to std::vector::resize in C#
Resize(classCandidates, numClass);
// find class id with highest score
var label = 0;
var score = 0.0f;
for (var j = 0; j < numClass; j++)
{
var classScore = clsProb[j];
if (classScore > score)
{
label = j;
score = classScore;
}
}
// ignore background or low score
if (label == 0 || score <= confidenceThresh)
{
continue;
}
// fprintf(stderr, "%d = %f\n", label, score);
// unscale to image size
var x1 = roi[0] / scale;
var y1 = roi[1] / scale;
var x2 = roi[2] / scale;
var y2 = roi[3] / scale;
var pbW = x2 - x1 + 1;
var pbH = y2 - y1 + 1;
// apply bbox regression
var dx = bboxPred[label * 4];
var dy = bboxPred[label * 4 + 1];
var dw = bboxPred[label * 4 + 2];
var dh = bboxPred[label * 4 + 3];
var cx = x1 + pbW * 0.5f;
var cy = y1 + pbH * 0.5f;
var objCx = cx + pbW * dx;
var objCy = cy + pbH * dy;
var objW = pbW * Math.Exp(dw);
var objH = pbH * Math.Exp(dh);
var objX1 = (float)(objCx - objW * 0.5f);
var objY1 = (float)(objCy - objH * 0.5f);
var objX2 = (float)(objCx + objW * 0.5f);
var objY2 = (float)(objCy + objH * 0.5f);
// clip
objX1 = Math.Max(Math.Min(objX1, bgr.Cols - 1), 0.0f);
objY1 = Math.Max(Math.Min(objY1, bgr.Rows - 1), 0.0f);
objX2 = Math.Max(Math.Min(objX2, bgr.Cols - 1), 0.0f);
objY2 = Math.Max(Math.Min(objY2, bgr.Rows - 1), 0.0f);
// append object
var obj = new Object
{
Rect = new Rect <float>(objX1, objY1, objX2 - objX1 + 1, objY2 - objY1 + 1),
Label = label,
Prob = score
};
classCandidates[label].Add(obj);
}
// post process
objects.Clear();
for (var i = 0; i < (int)classCandidates.Count; i++)
{
var candidates = classCandidates[i];
QsortDescentInplace(candidates);
var picked = new List <int>();
NmsSortedBBoxes(candidates, picked, nmsThreshold);
for (var j = 0; j < picked.Count; j++)
{
var z = picked[j];
objects.Add(candidates[z]);
}
}
QsortDescentInplace(objects);
if (maxPerImage > 0 && maxPerImage < objects.Count)
{
Resize(objects, maxPerImage);
}
}
return(0);
}
private static int DetectPoseNet(NcnnDotNet.OpenCV.Mat bgr, List <KeyPoint> keyPoints)
{
using (var poseNet = new Net())
{
if (Ncnn.IsSupportVulkan)
{
poseNet.Opt.UseVulkanCompute = true;
}
// the simple baseline human pose estimation from gluon-cv
// https://gluon-cv.mxnet.io/build/examples_pose/demo_simple_pose.html
// mxnet model exported via
// pose_net.hybridize()
// pose_net.export('pose')
// then mxnet2ncnn
// the ncnn model https://github.com/nihui/ncnn-assets/tree/master/models
poseNet.LoadParam("pose.param");
poseNet.LoadModel("pose.bin");
var w = bgr.Cols;
var h = bgr.Rows;
using var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr2Rgb, bgr.Cols, bgr.Rows, 192, 256);
// transforms.ToTensor(),
// transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)),
// R' = (R / 255 - 0.485) / 0.229 = (R - 0.485 * 255) / 0.229 / 255
// G' = (G / 255 - 0.456) / 0.224 = (G - 0.456 * 255) / 0.224 / 255
// B' = (B / 255 - 0.406) / 0.225 = (B - 0.406 * 255) / 0.225 / 255
var meanVals = new[] { 0.485f * 255.0f, 0.456f * 255.0f, 0.406f * 255.0f };
var normVals = new[] { 1 / 0.229f / 255.0f, 1 / 0.224f / 255.0f, 1 / 0.225f / 255.0f };
@in.SubstractMeanNormalize(meanVals, normVals);
using var ex = poseNet.CreateExtractor();
ex.Input("data", @in);
using var @out = new Mat();
ex.Extract("conv3_fwd", @out);
// resolve point from heatmap
keyPoints.Clear();
for (var p = 0; p < @out.H; p++)
{
using var m = @out.Channel(p);
var maxProb = 0f;
var maxX = 0;
var maxY = 0;
for (var y = 0; y < @out.H; y++)
{
var ptr = m.Row(y);
for (var x = 0; x < @out.W; x++)
{
var prob = ptr[x];
if (prob > maxProb)
{
maxProb = prob;
maxX = x;
maxY = y;
}
}
}
var keyPoint = new KeyPoint
{
P = new Point <float>(maxX * w / (float)@out.W, maxY * h / (float)@out.H),
Prob = maxProb
};
keyPoints.Add(keyPoint);
}
}
return(0);
}
private static int DetectRFCN(NcnnDotNet.OpenCV.Mat bgr, List <Object> objects)
{
using (var rfcn = new Net())
{
if (Ncnn.IsSupportVulkan)
{
rfcn.Opt.UseVulkanCompute = true;
}
// original pretrained model from https://github.com/YuwenXiong/py-R-FCN
// https://github.com/YuwenXiong/py-R-FCN/blob/master/models/pascal_voc/ResNet-50/rfcn_end2end/test_agnostic.prototxt
// https://1drv.ms/u/s!AoN7vygOjLIQqUWHpY67oaC7mopf
// resnet50_rfcn_final.caffemodel
rfcn.LoadParam("rfcn_end2end.param");
rfcn.LoadModel("rfcn_end2end.bin");
const int targetSize = 224;
const int maxPerImage = 100;
const float confidenceThresh = 0.6f; // CONF_THRESH
const float nmsThreshold = 0.3f; // NMS_THRESH
// scale to target detect size
var w = bgr.Cols;
var h = bgr.Rows;
float scale;
if (w < h)
{
scale = (float)targetSize / w;
w = targetSize;
h = (int)(h * scale);
}
else
{
scale = (float)targetSize / h;
h = targetSize;
w = (int)(w * scale);
}
using var @in = Mat.FromPixelsResize(bgr.Data, PixelType.Bgr, bgr.Cols, bgr.Rows, w, h);
var meanVals = new[] { 102.9801f, 115.9465f, 122.7717f };
@in.SubstractMeanNormalize(meanVals, null);
using var im_info = new Mat(3);
im_info[0] = h;
im_info[1] = w;
im_info[2] = scale;
// step1, extract feature and all rois
using var ex1 = rfcn.CreateExtractor();
ex1.Input("data", @in);
ex1.Input("im_info", im_info);
using var rfcnCls = new Mat();
using var rfcnBBox = new Mat();
using var rois = new Mat();// all rois
ex1.Extract("rfcn_cls", rfcnCls);
ex1.Extract("rfcn_bbox", rfcnBBox);
ex1.Extract("rois", rois);
// step2, extract bbox and score for each roi
var classCandidates = new List <List <Object> >();
for (var i = 0; i < rois.C; i++)
{
using var ex2 = rfcn.CreateExtractor();
using var roi = rois.Channel(i); // get single roi
ex2.Input("rfcn_cls", rfcnCls);
ex2.Input("rfcn_bbox", rfcnBBox);
ex2.Input("rois", roi);
using var bboxPred = new Mat();
using var clsProb = new Mat();
ex2.Extract("bbox_pred", bboxPred);
ex2.Extract("cls_prob", clsProb);
var numClass = clsProb.W;
// There is no equivalent to std::vector::resize in C#
Resize(classCandidates, numClass);
// find class id with highest score
var label = 0;
var score = 0.0f;
for (var j = 0; j < numClass; j++)
{
var classScore = clsProb[j];
if (classScore > score)
{
label = j;
score = classScore;
}
}
// ignore background or low score
if (label == 0 || score <= confidenceThresh)
{
continue;
}
// fprintf(stderr, "%d = %f\n", label, score);
// unscale to image size
var x1 = roi[0] / scale;
var y1 = roi[1] / scale;
var x2 = roi[2] / scale;
var y2 = roi[3] / scale;
var pbW = x2 - x1 + 1;
var pbH = y2 - y1 + 1;
// apply bbox regression
var dx = bboxPred[4];
var dy = bboxPred[4 + 1];
var dw = bboxPred[4 + 2];
var dh = bboxPred[4 + 3];
var cx = x1 + pbW * 0.5f;
var cy = y1 + pbH * 0.5f;
var objCx = cx + pbW * dx;
var objCy = cy + pbH * dy;
var objW = pbW * Math.Exp(dw);
var objH = pbH * Math.Exp(dh);
var objX1 = (float)(objCx - objW * 0.5f);
var objY1 = (float)(objCy - objH * 0.5f);
var objX2 = (float)(objCx + objW * 0.5f);
var objY2 = (float)(objCy + objH * 0.5f);
// clip
objX1 = Math.Max(Math.Min(objX1, bgr.Cols - 1), 0.0f);
objY1 = Math.Max(Math.Min(objY1, bgr.Rows - 1), 0.0f);
objX2 = Math.Max(Math.Min(objX2, bgr.Cols - 1), 0.0f);
objY2 = Math.Max(Math.Min(objY2, bgr.Rows - 1), 0.0f);
// append object
var obj = new Object
{
Rect = new Rect <float>(objX1, objY1, objX2 - objX1 + 1, objY2 - objY1 + 1),
Label = label,
Prob = score
};
classCandidates[label].Add(obj);
}
// post process
objects.Clear();
for (var i = 0; i < (int)classCandidates.Count; i++)
{
var candidates = classCandidates[i];
QsortDescentInplace(candidates);
var picked = new List <int>();
NmsSortedBBoxes(candidates, picked, nmsThreshold);
for (var j = 0; j < picked.Count; j++)
{
var z = picked[j];
objects.Add(candidates[z]);
}
}
QsortDescentInplace(objects);
if (maxPerImage > 0 && maxPerImage < objects.Count)
{
Resize(objects, maxPerImage);
}
}
return(0);
}