ncnn出问题

[tututu11]

运行ncnn例子，无任何修改，出现这种情况，哪位大佬知道是什么问题

[chatop2020]

ncnn的c++ demo上，我感觉有些问题，我用.net 重写一下，效果还行，你也可以自己改改，主要代码如下：

 outImg = extractor.Extract("758");
                if (outImg != null)
                {
                    for (int h = 0; h < outImg.H; h++)
                    {
                        for (int w = 0; w < outImg.H; w++)
/*
####这里顺便问一下大佬，outImg.H 真的对吗？是不是 你c++代码里笔误了，应该是outImg.W?#####
####我试过outImg.W好像和outImg.H没什么区别######################################
####其实我后来想想，也无所谓，因为进来的图片都是长宽一样的，可能就是因为这个，所以才无所谓吧######
*/
                        {
                            //前景概率
                            int obj_score_index = (0 * outImg.H * outImg.W) + (h * outImg.W) + w;
                            float obj_score = outImg[obj_score_index];
                            //解析类别
                            int category = 0;
                            float max_score = 0.0f;
                            for (long i = 0; i < _namesDic.Count; i++)
                            {
                                obj_score_index = (int) ((5 + i) * outImg.H * outImg.W) + (h * outImg.W) + w;
                                float cls_score = outImg[obj_score_index];
                                if (cls_score > max_score)
                                {
                                    max_score = cls_score;
                                    category = (int) i;
                                }
                            }

                            float score = (float) (Math.Pow(max_score, 0.4) * Math.Pow(obj_score, 0.6));
                            // 解析坐标
                            int x_offset_index = (1 * outImg.H * outImg.W) + (h * outImg.W) + w;
                            int y_offset_index = (2 * outImg.H * outImg.W) + (h * outImg.W) + w;
                            int box_width_index = (3 * outImg.H * outImg.W) + (h * outImg.W) + w;
                            int box_height_index = (4 * outImg.H * outImg.W) + (h * outImg.W) + w;


                            float x_offset = Tanh(outImg[x_offset_index]);
                            float y_offset = Tanh(outImg[y_offset_index]);
                            float box_width = Sigmoid(outImg[box_width_index]);
                            float box_height = Sigmoid(outImg[box_height_index]);

                            float cx = (w + x_offset) / outImg.W;
                            float cy = (h + y_offset) / outImg.H;

                            int x1 = (int) ((cx - box_width * 0.5) * inputImg.Cols);
                            int y1 = (int) ((cy - box_height * 0.5) * inputImg.Rows);
                            int x2 = (int) ((cx + box_width * 0.5) * inputImg.Cols);
                            int y2 = (int) ((cy + box_height * 0.5) * inputImg.Rows);
                            var box = new BoundingBox();
                            box.prob = score;
                            box.h = (uint) (y2 - y1);
                            box.w = (uint) (x2 - x1);
                            box.x = (uint) x1;
                            box.y = (uint) y1;
                            box.obj_id = (uint) category;
                            if (score > threshold && x1 >= 0 && y1 >= 0 && x2 <= inputImg.Cols && y2 <= inputImg.Rows)
                            {
                                outBBoxEx.FilterBoundingBoxes.Add(box);
                            }
                            else
                            {
                                outBBoxEx.OriginalBoundingBoxes.Add(box);
                            }
                        }
                    }

                    if (outBBoxEx.FilterBoundingBoxes.Count > 0)
                    {
                        outBBoxEx.FilterBoundingBoxes.Sort((x, y) => -x.prob.CompareTo(y.prob));
                        List<picks> picked = new List<picks>();
                        for (int i = 0; i < outBBoxEx.FilterBoundingBoxes.Count; i++)
                        {
                            bool keep = true;
                            for (int j = 0; j < picked.Count; j++)
                            {
                                if (picked[j].Remove) //如果是被过滤掉的，就不执行后续操作
                                {
                                    continue;
                                }

                                var obj1 = outBBoxEx.FilterBoundingBoxes[i];
                                var obj2 = outBBoxEx.FilterBoundingBoxes[picked[j].Index];
                                var rect1 = new Rectangle((int) obj1.x, (int) obj1.y, (int) obj1.w, (int) obj1.h);
                                var rect2 = new Rectangle((int) obj2.x, (int) obj2.y, (int) obj2.w, (int) obj2.h);
                                var a = bbOverlap(rect1, rect2); //计算两个矩形的交集，如果大于65%的就过滤掉
                                if (a > 0.65f && obj1.obj_id == obj2.obj_id)
                                {
                                    keep = false;
                                    break;
                                }

                                if (rect2.IntersectsWith(rect1)) //计算两个矩形的并集，如果是包含关系，同时obj1的分数大于obj2,还是同一个类型的，就保留当前的obj1
                                {
                                    if (obj1.prob > obj2.prob && obj1.obj_id == obj2.obj_id)
                                    {
                                        picked[j].Remove = true;
                                        break;
                                    }

                                    keep = false;
                                    break;
                                }
                            }

                            if (keep)
                            {
                                var p = new picks();
                                p.Index = i;
                                p.Remove = false;
                                picked.Add(p);
                            }
                        }

                        for (int i = 0; i < picked.Count; i++)
                        {
                            if (picked[i].Remove)
                            {
                                continue;
                            }

                            var obj = outBBoxEx.FilterBoundingBoxes[picked[i].Index];
                            if (obj.x >= 0 && obj.y >= 0 && obj.w <= inputImg.Cols && obj.h <= inputImg.Rows)
                                outBBoxEx.FinishBoundingBoxes.Add(obj);
                        }
                        outBBoxEx.FilterBoundingBoxes.Clear();
                        outBBoxEx.FilterBoundingBoxes = outBBoxEx.FinishBoundingBoxes;

                        Bitmap img = new Bitmap(filePath);
                        outBBoxEx.FilePath = filePath;
                        outBBoxEx.IsImgType = img.RawFormat;
                        outBBoxEx.Height = inputImg.Rows;
                        outBBoxEx.Width = inputImg.Cols;
                        outBBoxEx.PixelFormat = img.PixelFormat;
                        outBBoxEx.Threshold = threshold;
                        outBBoxEx.DetectTime = DateTime.Now;
                        FileInfo fi = new FileInfo(filePath);
                        outBBoxEx.FileSize = (ulong) fi.Length;
                        if (createResultImage)
                        {
                            darwBox(outBBoxEx, img, filePath + ".jpg");
                            outBBoxEx.ResultPath = filePath + ".jpg";
                        }

                        outBBoxEx.ElapsedMsc = (long) (DateTime.Now - dt1).TotalMilliseconds;
                        return outBBoxEx;
                    }
                }

[clue3clue32]

你是在vs上部署的吗？问题在于带“阈值”两个字的注释，注释下一行的代码出问题了。我是删了注释或者后面加个空格解决可以解决。