Rekognition.detectLabels: Face Label missing instances

0

Hello,

What is the meaning of the below output from Rekognition.detectLabels (Management Console) when the picture contains many clear, big faces, but the AI detects no instance ? Does the instance array contains the detected faces, or does it have another meaning (where can I found the specification for the Rekognition.detectLabels.DetectLabelsResult) ? On same picture Rekognition.indexFaces detects some faces.
In SDK the specification for Rekognition.detectLabels.DetectLabelsResult.getLabels.getInstances is:

If Label represents an object, Instances contains the bounding boxes for each instanceof the detected object. Bounding boxes are returned for common object labels such as people, cars, furniture, apparel or pets.

So, shouldn't the instances contains bounding boxes, as long as Face is a bounding box Label ?

		{
			"name": "Face",
			"aliases": [],
			"parents": [
				{
					"name": "Head"
				},
				{
					"name": "Person"
				}
			],
			"instances": [],
			"categories": [
				{
					"name": "Person Description"
				}
			],
			"confidence": 99.99898
		}

Thank you,
Mihai ADAM

1 回答
0

Hello,

After some research, I found out some reasons for the missing of the Face label instances: the DetectLabels method does not count the faces with low quality pixels (blurry, dark, extreme pose, few details, too small), but when the Confidence that faces exists is high, the label appears even without instances. More details could be found in this post

Have a good day,
Mihai

已回答 7 个月前

您未登录。 登录 发布回答。

一个好的回答可以清楚地解答问题和提供建设性反馈,并能促进提问者的职业发展。

回答问题的准则

相关内容