You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: articles/ai-services/cognitive-services-limited-access.md
+3-3
Original file line number
Diff line number
Diff line change
@@ -6,7 +6,7 @@ author: PatrickFarley
6
6
manager: nitinme
7
7
ms.service: azure-ai-services
8
8
ms.topic: conceptual
9
-
ms.date: 09/18/2024
9
+
ms.date: 03/26/2025
10
10
ms.author: pafarley
11
11
---
12
12
@@ -27,7 +27,7 @@ The following services are Limited Access:
27
27
-[Custom Neural Voice](/legal/cognitive-services/speech-service/text-to-speech/limited-access?context=/azure/ai-services/speech-service/context/context): Pro features
28
28
-[Custom text to speech avatar](/legal/cognitive-services/speech-service/text-to-speech/limited-access?context=/azure/ai-services/speech-service/context/context): All features
29
29
-[Speaker Recognition](/legal/cognitive-services/speech-service/speaker-recognition/limited-access-speaker-recognition?context=/azure/ai-services/speech-service/context/context): All features
30
-
-[Face API](/legal/cognitive-services/computer-vision/limited-access-identity?context=/azure/ai-services/computer-vision/context/context): Identify and Verify features, face ID property
30
+
-[Azure AI Face](/legal/cognitive-services/computer-vision/limited-access-identity?context=/azure/ai-services/computer-vision/context/context): Identify and Verify features, face ID property
31
31
-[Azure AI Vision](/legal/cognitive-services/computer-vision/limited-access?context=/azure/ai-services/computer-vision/context/context): Celebrity Recognition feature
32
32
-[Azure AI Video Indexer](/azure/azure-video-indexer/limited-access-features): Celebrity Recognition and Face Identify features
> Face attributes are predicted through the use of statistical algorithms. They might not always be accurate. Use caution when you make decisions based on attribute data. Please refrain from using these attributes for anti-spoofing. Instead, we recommend using Face Liveness detection. For more information, please refer to [Tutorial: Detect liveness in faces](/azure/ai-services/computer-vision/tutorials/liveness).
23
+
> Face attributes are predicted by statistical algorithms. They might not always be accurate. Use caution when you make decisions based on attribute data. Refrain from using these attributes for anti-spoofing. Instead, we recommend using Face Liveness detection. For more information, please refer to [Tutorial: Detect liveness in faces](/azure/ai-services/computer-vision/tutorials/liveness).
24
24
25
25
This article explains the concepts of face detection and face attribute data. Face detection is the process of locating human faces in an image and optionally returning different kinds of face-related data.
26
26
27
-
You use the [Detect] API to detect faces in an image. To get started using the REST API or a client SDK, follow a [Face service quickstart](./quickstarts-sdk/identity-client-library.md). Or, for a more in-depth guide, see [Call the detect API](./how-to/identity-detect-faces.md).
27
+
You use the [Detect] API to detect faces in an image. To get started using the REST API or a client SDK, follow a [Face service quickstart](./quickstarts-sdk/identity-client-library.md). Or, for a more in-depth guide, see [Call the Detect API](./how-to/identity-detect-faces.md).
Copy file name to clipboardExpand all lines: articles/ai-services/computer-vision/concept-object-detection.md
+5-3
Original file line number
Diff line number
Diff line change
@@ -7,15 +7,17 @@ manager: nitinme
7
7
8
8
ms.service: azure-ai-vision
9
9
ms.topic: conceptual
10
-
ms.date: 09/19/2024
10
+
ms.date: 03/26/2025
11
11
ms.author: pafarley
12
12
---
13
13
14
14
# Object detection
15
15
16
-
This article explains the concept of object detection. Object detection is similar to [tagging](concept-tag-images-40.md), but the API returns the bounding box coordinates (in pixels) for each object found in the image. For example, if an image contains a dog, cat, and person, the object detection operation lists those objects with their coordinates in the image. You can use this functionality to process the relationships between the objects in an image. It also lets you determine whether there are multiple instances of the same object in an image.
16
+
This article explains the object detection feature. Object detection is similar to [tagging](concept-tag-images-40.md), but the API returns the bounding box coordinates (in pixels) for each object found in the image. For example, if an image contains a dog, cat, and person, the object detection operation lists those objects with their coordinates in the image.
17
17
18
-
The object detection function applies tags based on the objects or living things identified in the image. There's no formal relationship between the tagging taxonomy and the object detection taxonomy. At a conceptual level, the object detection function only finds objects and living things, while the tag function can also include contextual terms like *indoor*, which can't be localized with bounding boxes.
18
+
You can use this functionality to process the relationships between the objects in an image. It also lets you determine whether there are multiple instances of the same object in an image.
19
+
20
+
There's no formal relationship between the tagging taxonomy and the object detection taxonomy. At a conceptual level, the object detection function only finds objects and living things, while the tag function can also include contextual terms like *indoor*, which can't be localized with bounding boxes.
19
21
20
22
Try out the capabilities of object detection quickly and easily in your browser by using Azure AI Vision Studio.
Copy file name to clipboardExpand all lines: articles/ai-services/content-safety/how-to/containers/container-overview.md
+3-3
Original file line number
Diff line number
Diff line change
@@ -6,7 +6,7 @@ author: PatrickFarley
6
6
manager: nitinme
7
7
ms.service: azure-ai-content-safety
8
8
ms.topic: overview
9
-
ms.date: 09/23/2024
9
+
ms.date: 03/26/2025
10
10
ms.author: pafarley
11
11
keywords: on-premises, Docker, container
12
12
---
@@ -17,7 +17,7 @@ Containers let you use a subset of the Azure AI Content Safety features in your
17
17
18
18
## Available containers
19
19
20
-
The following table lists the content safety containers available in the Microsoft Container Registry (MCR). The table also lists the features supported by each container and the latest version of the container.
20
+
The following table lists the content safety containers available in the Microsoft Container Registry (MCR). The table also lists the features supported by each container and the latest version of the container.
Copy file name to clipboardExpand all lines: articles/ai-services/content-safety/how-to/embedded-content-safety.md
+26-29
Original file line number
Diff line number
Diff line change
@@ -6,97 +6,94 @@ author: PatrickFarley
6
6
manager: nitinme
7
7
ms.service: azure-ai-content-safety
8
8
ms.topic: how-to
9
-
ms.date: 9/24/2024
10
-
ms.author: zhanxia
9
+
ms.date: 03/26/2025
10
+
ms.author: pafarley
11
11
---
12
12
13
13
# Embedded content safety (preview)
14
14
15
-
Embedded content safety is designed for on-device scenarios where cloud connectivity is intermittent or prefer on-device for privacy reason. For example, you can use embedded content safety in a PC to detect harmful content generated by foundation model, or a car that might travel out of range. You can also develop hybrid cloud and offline solutions. For scenarios where your devices must be in a secure environment like a bank or government entity, you should first consider [disconnected containers](../../containers/disconnected-containers.md).
15
+
Embedded content safety is designed for on-device scenarios where cloud connectivity is intermittent or the user prefers on-device access for privacy reasons.
16
+
17
+
You can use embedded content safety locally on a PC to detect harmful content generated by a large language model, or in a car that might travel out of a specified range. You can also develop hybrid cloud and offline solutions. For scenarios where your devices must be in a secure environment like a bank or government entity, you should first consider [disconnected containers](../../containers/disconnected-containers.md).
16
18
17
19
> [!IMPORTANT]
18
-
> Microsoft limits access to embedded content safety. You can apply for access through the Azure AI content safety [embedded content safety limited access review](https://aka.ms/aacs-embedded-application). For more information, see [Limited access](../limited-access.md).
20
+
> Microsoft limits access to embedded content safety. You can apply for access through the Azure AI content safety [embedded content safety limited access review](https://aka.ms/aacs-embedded-application). Instructions are provided upon successful completion of the limited access review process. For more information, see [Limited access](../limited-access.md).
19
21
20
22
## Platform requirements
21
23
22
-
Embedded content safety is included with the content safety C++ SDK.
23
-
24
-
**Choose your target environment**
24
+
Embedded content safety is included with the Azure AI Content Safety C++ SDK.
25
25
26
-
Embedded content safety only supports Windows right now. Contact your Microsoft account contact if you need to run embedded content safety on a different platform.
26
+
### Choose your target environment
27
27
28
-
# [Windows X64](#tab/windows-target)
28
+
Embedded content safety only supports Windows. Contact your Microsoft account administrator if you need to run embedded content safety on a different platform.
29
29
30
30
Requires Windows 10 or newer on x64 hardware.
31
31
32
32
The latest [Microsoft Visual C++ Redistributable for Visual Studio 2015-2022](/cpp/windows/latest-supported-vc-redist?view=msvc-170&preserve-view=true) must be installed regardless of the programming language used with the content safety SDK.
33
33
34
-
---
35
34
36
35
## Limitations
37
36
38
-
Embedded content safety is only available with C++ SDK. The other content safety SDKs, and REST APIs don't support embedded content safety.
37
+
Embedded content safety is only available with the C++ SDK. The other Content Safety SDKs and REST APIs don't support embedded content safety.
39
38
40
39
41
40
## Embedded content safety SDK packages
42
41
43
42
44
-
For C++ embedded applications, install following content safety SDK for C++ packages:
43
+
For C++ embedded applications, install the following C++ packages:
45
44
46
45
|Package |Description |
47
46
| --------- | --------- |
48
47
|[Azure.AI.ContentSafety.Extension.Embedded.Text](https://www.nuget.org/packages/Azure.AI.ContentSafety.Extension.Embedded.Text)|Required to run text analysis on device|
49
48
|[Azure.AI.ContentSafety.Extension.Embedded.Image](https://www.nuget.org/packages/Azure.AI.ContentSafety.Extension.Embedded.Image)|Required to run image analysis on device|
50
49
51
50
52
-
53
-
54
51
## Models
55
52
56
-
For embedded content safety, you need to download the content safety to your device. Microsoft limits access to embedded content safety. You can apply for access through the [embedded content safety limited access review](https://aka.ms/aacs-embedded-application). Instructions are provided upon successful completion of the limited access review process.
53
+
For embedded content safety, you need to download the content safety to your device.
57
54
58
-
The embedded content safety supports [analyze text](../quickstart-text.md) and [analyze image](../quickstart-image.md) features. These features scan text or image content for sexual content, violence, hate, and self-harm with multiple severity levels. It should be noted that these embedded models have been optimized for on-device execution with less computational resources compared to the Azure API. Therefore, it's possible that the output generated from the embedded content safety model may vary from that of the Azure API.
55
+
The embedded content safety supports [Analyze text](../quickstart-text.md) and [Analyze image](../quickstart-image.md) features. These features scan text or image content for sexual content, violence, hate, and self-harm with multiple severity levels.
59
56
57
+
These embedded models have been optimized for on-device execution with less computational resources compared to the Azure API. Therefore, it's possible that the output generated from the embedded content safety model may vary from that of the Azure API.
60
58
61
-
## Embedded content safety code samples
62
59
63
-
Below is the ready to use embedded content safety samples. Follow the readme file to run the sample.
Embedded content safety models run fully on your target devices. Understanding the performance characteristics of these models on your devices’ hardware can be critical to delivering low latency experiences within your products and applications. This section provides information to help answer the question, "Is my device suitable to run embedded content safety for text analysis or image analysis?"
69
+
Embedded content safety models run fully on your target devices. Understanding the performance characteristics of these models on your devices' hardware can be critical to delivering low latency experiences within your products and applications. This section provides information to help determine if your device is suitable to run embedded content safety for text analysis or image analysis.
72
70
73
71
### Factors that affect performance
74
-
Device specifications – The specifications of your device play a key role in whether embedded content safety models can run without performance issues. CPU clock speed, architecture (for example, x64, ARM processor, etcetera), and memory can all affect model inference speed.
75
72
76
-
CPU/GPU load – In most cases, your device is running other applications in parallel to the application where embedded content safety models are integrated. The amount of CPU/GPU load your device experiences when idle and at peak can also affect performance.
73
+
**Device specifications** – The specifications of your device play a key role in whether embedded content safety models can run without performance issues. CPU clock speed, architecture (for example, x64, ARM processor, etcetera), and memory can all affect model inference speed.
74
+
75
+
**CPU/GPU load** – In most cases, your device is running other applications in parallel to the application where embedded content safety models are integrated. The amount of CPU/GPU load your device experiences when idle and at peak can also affect performance.
77
76
78
77
For example, if the device is under moderate to high CPU load from all other applications running on the device, it's possible to encounter performance issues for running embedded content safety in addition to the other applications, even with a powerful processor.
79
78
80
-
Memory load – An embedded content safety text analysis process consumes about 900 MB of memory at runtime. If your device has less memory available for the embedded content safety process to use, frequent fallbacks to virtual memory and paging can introduce more latencies. This can affect both the real-time factor and user-perceived latency.
79
+
**Memory load** – An embedded content safety text analysis process consumes about 900 MB of memory at runtime. If your device has less memory available for the embedded content safety process to use, frequent fallbacks to virtual memory and paging can introduce more latencies. This can affect both the real-time factor and user-perceived latency.
81
80
82
-
### SDK parameters that can impact performance
81
+
### SDK parameters that can affect performance
83
82
84
83
The following SDK parameters can impact the inference time of the embedded content safety model.
85
84
86
85
-`gpuEnabled` Set as **true** to enable GPU, otherwise CPU is used. Generally inference time is shorter on GPU.
87
86
-`numThreads` This parameter only works for CPU. It defines number of threads to be used in a multi-threaded environment. We support a maximum number of four threads.
88
87
89
-
See next section for performance benchmark data on popular PC CPUs and GPUs.
90
-
91
88
92
89
### Performance benchmark data on popular CPUs and GPUs
93
90
94
-
As stated above, there are multiple factors that impact the performance of embedded content safety model. We highly suggest you test it on your device and tweak the parameters to fit for your application's requirement.
91
+
As stated above, there are multiple factors that impact the performance of an embedded content safety model. We highly recommend you test it on your device and tweak the parameters to fit for your application's requirements.
95
92
96
93
We also conduct performance benchmark tests on various popular PC CPUs and GPUs. Keep in mind that even with the same CPU, performance can vary depending on the CPU and memory load. The benchmark data provided should serve as a reference when considering if the embedded content safety can operate on your device. For optimal results, we advise testing on your intended device and in your specific application scenario.
97
94
98
95
99
-
The [sample code](https://github.com/Azure/azure-ai-content-safety-sdk) includes code snippet to monitor performance metrics like memory, inference time.
96
+
The [sample code](#code-samples) includes code snippets to monitor performance metrics like memory, inference time.
100
97
101
98
#### [Text analysis performance](#tab/text)
102
99
@@ -166,6 +163,6 @@ The [sample code](https://github.com/Azure/azure-ai-content-safety-sdk) includes
166
163
167
164
---
168
165
169
-
## Related Content
166
+
## Related content
170
167
171
168
-[Limited access to Content Safety](../limited-access.md)
0 commit comments