简体繁体 English

为什么object检测会发现多个物体？

[英]Why does object detection result in multiple found objects?

原文 2020-07-09 15:55:31 2 1 ios/ machine-learning/ object-detection/ coreml/ createml

I trained an object detector with CreateML and when I test the model in CreateML, I get a high number of identified objects:我用 CreateML 训练了一个 object 检测器，当我在 CreateML 中测试 model 时，我得到了大量的识别对象：

Notes:笔记：

The model was trained on a small data set of ~30 images with that particular label face-gendermale occuring ~20 times. model 在一个包含约 30 张图像的小型数据集上进行了训练，其中特定的 label face-gendermale出现了约 20 次。
Each training image has 1-3 labelled objects.每个训练图像有 1-3 个标记对象。
There are 5 label total.总共有 5 个 label。

Questions:问题：

Is that expected or is there something wrong with the model?这是预期的还是 model 有问题？
If this is expected, how should I evaluate these multiple results or even count the number of objects found in the model?如果这是意料之中的，我应该如何评估这些多个结果，甚至计算在 model 中找到的对象数量？

Cross-posted in Apple Developer Forums .在Apple 开发者论坛中交叉发布。 Photo of man © Jason Stitt |男人的照片© Jason Stitt | Dreamstime.com Dreamstime.com

1 个解决方案

A typical object detection model with make about 1000 predictions for every image (although it can be much more depending on the model architecture).典型的 object 检测 model 对每张图像进行大约 1000 次预测（尽管它可能更多取决于 model 架构）。 Most of these predictions have very low confidence, so they are filtered out.大多数这些预测的置信度都非常低，因此被过滤掉了。 Then the ones that are left over are sent through non-maximum suppression (NMS), which removes bounding boxes that overlap too much.然后剩下的那些通过非最大抑制（NMS）发送，它去除了重叠太多的边界框。

In your case, it seems that the threshold for NMS is too low (or too high), because many overlapping boxes survive.在您的情况下，NMS 的阈值似乎太低（或太高），因为许多重叠框仍然存在。

However, it also seems that the model hasn't been trained very well yet, probably because you used very few images.但是，model 似乎还没有训练得很好，可能是因为您使用的图像很少。