Skip to content

release-notes.md

Latest commit

 

History

History
523 lines (365 loc) · 32.1 KB

release-notes.md

File metadata and controls

523 lines (365 loc) · 32.1 KB
copyright
years
2015, 2019
lastupdated 2019-08-16
keywords new features,updates to Visual Recognition,what's new with Visual Recognition
subcollection visual-recognition

{:shortdesc: .shortdesc} {:external: target="_blank" .external} {:tip: .tip} {:important: .important} {:note: .note} {:deprecated: .deprecated} {:pre: .pre} {:codeblock: .codeblock} {:screen: .screen}

Release notes

{: #release-notes}

The following new features and changes to the service are available. {: shortdesc}

Service API Versioning

{: version}

API requests require a version parameter that takes a date in the format version=YYYY-MM-DD. Whenever we change the API in a backwards-incompatible way, we release a new minor version of the API.

Send the version parameter with every API request. The service uses the API version for the date you specify, or the most recent version before that date. Don't default to the current date. Instead, specify a date that matches a version that is compatible with your app, and don't change it until your app is ready for a later version.

The current version is 2018-03-19.

Beta features

{: #beta}

{{site.data.keyword.IBM_notm}} releases services, features, and language support for your evaluation that are classified as beta. These features might be unstable, might change frequently, and might be discontinued with short notice. Beta features also might not provide the same level of performance or compatibility that generally available features provide and are not intended for use in a production environment. Beta features are supported only on IBM Developer Answers{: external}.

Changes

{: #changelog}

The following new features and changes to the service are available.

16 August 2019

{: #16august2019}

  • Custom Object Detection avaiable in Beta
    • Custom Object Detection identifies items and their location in an image. The service detects these items based on a set of images with labeled training data that you provide. For details, see Custom Object Detection (Beta)
    • Custom Object Detection is supported by the Visual Recognition v4 API (Beta){: external}

21 March 2019

{: #21march2019}

  • Change to viewing service credentials

    • Users can now see service credential information only that is associated with the role assigned to your {{site.data.keyword.cloud_notm}} account. For example, if you are assigned a reader role, any writer or higher levels of service credentials are not visible.

      This change does not affect API access for users or applications with existing service key credentials. Only the viewing of credentials within {{site.data.keyword.cloud_notm}} is affected.

      For more information about service keys and user roles, see IAM service API keys.

15 January 2019

{: #15january2019}

  • Translation for gender in Detect Faces
    • The Detect Faces methods now return translated labels for "Male" and "Female" when you provide the language in the Accept-Language request header. For details see the API reference{: external}.

1 October 2018

{: #01october2018}

  • Service instances created before May 23, 2018 are deleted.

    • As previously notified, all {{site.data.keyword.visualrecognitionshort}} instances created before May 23, 2018 are no longer active. Data from the instances is now deleted. See Migrating for details about how to move to a new service instance.
    • Service instances created after this date are not affected.
    • If you have any questions, contact IBM support{: external}.

1 August 2018

{: #01august2018}

  • General availability of Food and Explicit models

    The Food and Explicit models moved from Beta to General availability (GA). The Food model recognizes more food and meals than the General (default) model. The Explicit model identifies what might be considered pornographic images.

    No code change are required. Both models are free under the Lite plan and cost $0.002 per image under the Standard plan.

    For more information, see Updates to Watson Visual Recognition{: external} in the Watson blog.

1 July 2018

{: #01july2018}

21 June 2018

{: #21june2018}

  • Additional language support

    • The Classify methods now support Chinese (Simplified and Traditional) and Portuguese (Brazilian) in the output of default (General) model classes. For the full list of languages, see Supported languages.
    • All languages are now also supported in the responses from the Food and Explicit models.

22 May 2018

{: #22may2018}

  • New API authentication process:

    You now authenticate with Identity and Access Management (IAM) at a new endpoint:

    • Use a different endpoint URL for new instances. The default endpoint is https:/gateway.watsonplatform.net/visual-recognition/api/. To find the URL for your service instance, check the credentials by clicking the instance from the {{site.data.keyword.cloud_notm}} Resource list{: external}.
    • Modify how you authenticate to the API. You provide either an IAM key or access token for your service instance. See Migrating for examples.

    For service instances created before May 23, 2018, the authentication process and endpoint have not changed. Authenticate by providing the api_key query parameter.

  • Updates to the Lite plan

    Lite plans created after May 22, 2018 are changing:

    • New Lite plan instances continue to be available after 30 days if you use them every month.
    • You can create and retrain two custom models under the new Lite plan.
    • Lite plans include up to 1,000 events a month. Each image that you send for classification, detection, or training is an event. Downloading a Core ML model doesn't count toward the event limit.

    If your needs exceed the Lite plan, update to a billable account. Explore{: external} the pricing plans.

  • Information security:

    We updated the documentation to include some new details about data privacy. Read the details in Information security.

12 April 2018

{: #12april2018}

  • Support for retraining a custom model on the Lite plan

    Under the Lite plan, you no longer have to delete and create another custom model when you want to update or retrain the model. You can now update a custom model as long as you remain under daily and monthly limits{: external} of the plan.

    If you need to have multiple models or multiple versions of the same model, update from the Lite plan to a billable account.

  • New minor version with support for CORS

    Version: 2018-03-19

    We now support Cross-Origin Resource Sharing (CORS) headers when you specify version=2018-03-19 in requests.

5 April 2018

{: #5april2018}

  • Fix for Face model score issue

    The range for the age score in the Face model is restored to the original range of 0 to 1. For details, see the Known issues for 2 April 2018.

  • New form-data parameters

    The Classify images and Detect faces in images methods support new form-data parameters. Classify images supports separate url, classifier_ids, threshold, and owners parameters, and Detect faces supports url.

    In the past, you encoded the values in a JSON string and passed the parameters form-data parameter. You can use the new method of passing these values in separate form-data parameters for all application development. For details, see the API reference{: external}.

2 April 2018

{: #2april2018}

  • General availability of enhanced Face model

    The updated face detection model is now in General availability (GA).

    This enhanced model uses broader training data sets for increased accuracy of facial detection for age and gender. For example, age predictions are improved by reducing the range between min and max. The fixed 9-year age range in the previous model is replaced with a dynamic and smaller range. The average age range is now 4.9 years.

    • Changes to the API

    Other differences between the enhanced and previous Face model: - The enhanced model supports .gif and .tif image formats. - The enhanced model supports larger file sizes: up to 10 MB for image files and up to 100 MB for .zip files. - The enhanced model does not include FaceIdentity information in the response. The identity information refers to the name of the person, score, and type_hierarchy knowledge graph.

    For details about the API, see the API reference{: external}.

    • Known issues

      • Form parameters cannot include the Content-Type. For example, this request fails to analyze the url parameter because it includes ;type=text/plain:

        curl -F url="https://example.com/images/prez.jpg;type=text/plain" \
"https://gateway-a.watsonplatform.net/visual-recognition/api/v3/detect_faces?api_key={api-key}&version=2016-05-20"

        {: pre}

      • The age score has a range of 0 to 0.3. We are working to restore the original range of 0 to 1.

    • Beta endpoint deprecated

    The beta endpoint at /v3/detect_faces_beta is deprecated and will not be accessible after May 17, 2018. Make sure that your requests point to /v3/detect_faces.

20 March 2018

{: #20march2018}

  • Integration with Apple Core ML

    {{site.data.keyword.visualrecognitionshort}} now includes support for the Apple Core ML model format. You can use a Core ML version of your {{site.data.keyword.visualrecognitionshort}} model on your iOS apps.

    Start developing: To get started developing with {{site.data.keyword.visualrecognitionshort}} and Core ML, check out these projects on GitHub:

  • Changes to the API

    The following backward-compatible changes to the API are included in the integration:

    • A new core_ml_enabled field that indicates whether a classifier model can be downloaded as a Core ML model. The field is returned in the response for calls to GET and POST /v3/classifiers and GET /v3/classifiers/{classifier_id}.
    • A new GET /v3/classifiers/{classifier_id}/core_ml_model method to download a Core ML model as an .mlmodel file. You can download Core ML model files for custom models created after March 19.
    • A new updated field with the latest training date of the model. The updated field matches either the retrained field or the created field.

    For details about the API changes for Core ML, see the API reference{: external}.

  • New tool available: Watson Studio

    Watson Studio{: external} is the new integrated environment that includes a replacement for the beta {{site.data.keyword.visualrecognitionshort}} tool. Watson Studio supports not only {{site.data.keyword.visualrecognitionshort}} but also many other {{site.data.keyword.cloud_notm}} services and resources. You can use Watson Studio with all your existing {{site.data.keyword.visualrecognitionshort}} instances and classifiers.

    Watson Studio provides a collaborative environment in the cloud. With Watson Studio, developers, subject matter experts, data scientists, and others can build and train {{site.data.keyword.visualrecognitionshort}} and other AI models. You can also use Watson Studio to access the built-in General and Face models.

    Watson studio also supports Core ML. You can download a Core ML model file for your custom model.

    Get started{: external} with Watson Studio.

  • Updated deep learning architecture for custom models

    {{site.data.keyword.visualrecognitionshort}} now uses a faster and more efficient deep learning network architecture for classification. The updated models also can differentiate more strongly between the top class and the rest of the classes. This approach might result in somewhat longer training times. The new architecture is used to train new custom models. When you retrain existing older models, the original architecture is used.

    The following example shows the differentiation with the new architecture:

    Man archery bow and arrow. Photo by Annie Spratt on Unsplash

    Original class and score Updated class and score
    Archery
    0.99     		archery
    0.9
    Auto Racing
    0.996398    		 biking
    0.004
    Biking
    0.0500174    		 fishing
    0.001
    Fishing
    0.11029    		 golf
    0.031
    Golf
    0.0980796    		 gymnastics
    0.029
    Gymnastics
    0.964391    		 judo
    0.021
    Judo
    0.339119    		 racing
    0.002
    Skating
    0.0393602    		 skating
    0.061
    Skiing
    0.0310527    		 skiing
    0.003
    Track and Field
    0.208147    		 track
    0.035

  • French language support

    The Classify methods now support French in the output of default model classes. For the full list of languages, see Supported languages.

23 February 2018

{: #23february2018}

  • Enhanced Face model available in beta

    An updated face detection model is available. This beta model uses broader training data sets for increased accuracy of facial detection for age and gender. For more information, see Increasing the Accuracy of IBM’s Watson {{site.data.keyword.visualrecognitionshort}} service{: external} and Mitigating Bias in AI Models{: external}.

    • You can view results of the updated model in the demo{: external}.
    • The beta model is available at /v3/detect_faces_beta.

    Differences between the beta and general availability (GA) models:

    • The beta model supports .gif and .tif image formats; this enhancement is expected to be applied to the GA model.
    • The beta model supports larger file sizes: up to 10 MB for image files and up to 100 MB for .zip files. This enhancement is expected to be applied to the GA model.
    • Beta face detection does not include FaceIdentity information in the response.
    • The beta model's POST request requires a non-empty filename. The GA Face model does not enforce this constraint.
    • The beta model's POST request supports a separate form parameter called url. The GA model encloses that information in the parameters JSON object. For details, see the API explorer{: external}.

  • Face identity deprecated

    The identity information in the response of the GA Face model is deprecated and will be removed from the API on April 2, 2018. The identity information refers to the name of the person, score, and type_hierarchy knowledge graph.

16 January 2018

{: #16january2018}

  • Lite account and plan replaces the Free plan

    A Lite account is free; no credit card is required. However, Lite plan service instances are deleted after 30 days.

    The maximum number of API calls that you can make with the Lite plan is slightly different from the Free plan. You can make a maximum of 7,500 API calls per month at 250 calls per day on the Lite plan.

    To update to a billable account when you have custom classifiers, create another service instance and re-create your custom classifiers.

11 December 2017

{: #11december2017}

  • Increased accuracy and output with the General model

    The General model, which contains several thousand tags, now detects more secondary objects and has improved scene detection. These improvements help recognize the less prominent aspects of an image. In addition, the average number of tags returned per image has increased to 10.

    The following image shows an example of the tags returned before the update and the additional tags that are now returned.

    Photograph of azalea plant
    Original tags Additional tags
    Tag: tree
    Score: 0.799
    Tag: flower
    Score: 0.792
    Tag: alpine azalea
    Score: 0.696
    Tag: plant
    Score: 0.868
    Tag: azalea
    Score: 0.617
    Tag: swamp azalea
    Score: 0.5
    Tag: reddish orange color
    Score: 0.1
  • New Explicit model available in beta

    The Explicit model, which launches in beta, classifies whether an image contains pornographic content and is inappropriate for general use. You can include the Explicit model with other models for combined analysis. For example, include both the `default` and `explicit` classifier IDs in your request to return image tags and whether the image contains explicit content.

    For details about the API call, see the **Classify images** method in the [API reference](https://{DomainName}/apidocs/visual-recognition/#classify-images){: external}.

  • Support for larger file sizes

    The Classify images methods now support image files up to 10 MB and .zip files up to 100 MB.

  • Array required when passing classifier IDs

    The API now enforces an array when you pass in `classifier_ids` as part of the **parameters** object in the **Classify images** method. Previously, you could pass a classifier ID as a string. For more information, see the parameters description and example file in the [API reference](https://{DomainName}/apidocs/visual-recognition/#classify-images){: external}.

8 September 2017

{: #8september2017}

30 June 2017

{: #30june2017}

  • Improved tagging: We increased the number of training images for the default classifier. That increase improve the ability to recognize accurately the overall ‘scene’ of an image. For details, see Further Enhancements for General Tagging Feature{: external}

  • Additional languages: The Classify images method now supports Korean, Italian, and German in addition to English, Arabic, Spanish, and Japanese.

    For details about the API call, see the Classify an image method in the API reference{: external}.

16 May 2017

{: #16may2017}

  • New food classifier is available: Beta

    A new beta food recognition model provides enhanced specificity and accuracy for images of food items. The Classify an image method allows you to add this new classifier, "food", to the classifier_ids parameter.

5 April 2017

{: #5april2017}

  • New {{site.data.keyword.visualrecognitionshort}} tool is available: Beta

    A new beta feature, the {{site.data.keyword.visualrecognitionshort}} tool, is available. This tool helps you work more easily with the {{site.data.keyword.visualrecognitionshort}} service. By entering your {{site.data.keyword.cloud_notm}} API key, you can use a GUI to access General Tagging and Face Detection features, as well as to seamlessly create, retrain, and delete custom classifiers associated with your API key, without needing to code.

8 March 2017

{: #8march2017}

  • Updated language support for general classifier tagging

    The general classifier now returns tags in all supported languages.

  • Known issues

    FIXED 04-13-2017: Repeatedly calling GET /classifiers while training or retraining is in progress, in order to check the status, can result in killing the training job. To workaround this issue, poll a new classifier's status using GET /classifiers/{classifier_id}. In other words, use the classifier GET for a single classifier ID, rather than GET /classifiers, which gets all classifiers, and can trigger this issue. Note that GET /classifiers/{classifier_id} is also faster.

15 December 2016

{: #15december2016}

  • Improved general classifier tagging

    • The general classifier's deep learning algorithms have been updated. There is a significant improvement in tag quality for all languages, and a significant increase in the volume of English tags returned by the service. This high-quality tag output is also available when you create your own custom classifier.
    • Color tagging has been added. The service now returns the top one or two colors in the image.

  • Known issues

    • Images submitted to the demo with EXIF metadata tags do not specify orientation (landscape versus portrait) correctly to the service. iPhone uploaded images may contain EXIF metadata tags. The workaround for this is to update your image metadata to not rely on EXIF metadata tags to specify the orientation of your image. Often, you can do this just by opening the image on your computer and saving it. For example, on a Mac, opening in Preview and then saving the image sets the correct orientation tags.
    • Sending images to the {{site.data.keyword.visualrecognitionshort}} service with EXIF tag{: external} values of 8, 3 or 6 can add latency. The service rotates the pixels of the image to the encoded viewpoint. You can save time by pre-rotating your images, or by removing the EXIF headers if they are not important to your {{site.data.keyword.visualrecognitionshort}} task.

1 December 2016

{: #1december2016}

  • New pricing

    {{site.data.keyword.IBM_notm}} has lowered the pricing of custom classifiers on the {{site.data.keyword.visualrecognitionshort}} service and increased what's available on the free plan. For more information, see the {{site.data.keyword.visualrecognitionshort}} pricing page.

7 October 2016

{: #7october2016}

  • Face detection enhancements

    The service has a new face-detection algorithm that improves the responses of the GET and POST /v3/detect_faces methods. The service adjusted the filtering of low-confidence age, name, and gender facial detections. As a result, more faces are found, and a larger range of confidences are returned.

8 September 2016

{: #8september2016}

  • Similarity Search BETA

    Users can now upload their own collection of images, use an image to search that collection for similar images, and then the service returns the top 100 most similar images. Similarity search can be utilized for any purpose, and users can custom train the service on collections of up to 1 million images each.

  • Text recognition is now closed beta

    The POST and GET /v3/recognize_text methods have gone back into closed beta. {{site.data.keyword.IBM_notm}} looks forward to continuing to support BETA clients that use the service, with no current plans for another open beta.

1 August 2016

  • Introductory pricing ending

    Custom classifier training and retraining, custom image classification, and custom classifier storage is no longer free. For information on the pricing, see the {{site.data.keyword.visualrecognitionshort}} service pricing page{: external}.

5 July 2016

{: #5july2016}

  • Updating custom classifiers

    Custom classifiers are no longer limited to a fixed set of training data. A user can now update existing classifiers with new images, or add additional positive or negative example classifiers to an existing trained classifier. By supplying additional example images for Watson to learn from, the service can make a user's classifier more accurate. Custom classifiers can now learn over time, continuously getting better at understanding visual information.

  • Known issues

    FIXED 04-13-2017: Users may get a 500 error code after 30 or 90 seconds when updating an existing classifier. Despite the error, there is a good chance that the retrain request completes successfully. Wait at least 2 minutes, and then check the status of the classifier by using the GET classifiers/{classifier\_id} method. If the status is "ready" and the "retrained" timestamp has been updated, the retraining request which generated the 500 code was successful. Otherwise, if the "retrained" timestamp has not been updated, there is an explanation added to the response which says why the retraining failed, and the service will revert the classifier to the version there was before the retrain request was issued.

20 May 2016

{: #20 may 2016}

This is the General Availability release of the {{site.data.keyword.visualrecognitionshort}} service. This release introduces Version 3 of the service, and is a breaking change. This release incorporates functionality from the AlchemyVision service, which has been deprecated.

Any custom classifiers that were created while the service was in Beta must be recreated in a GA instance of the service.

The following changes and updates were made to the {{site.data.keyword.visualrecognitionshort}} service:

  • Version date: To utilize the features of this release, use 2016-05-20 as the value for the version parameter.
  • Classes and classifiers: Single classifiers are now called "classes". In GA, a group of classes is called a "classifier".
  • Classification: Use the POST or GET /v3/classify methods to quickly and accurately identify a variety of subjects and scenes with default classes.
  • Face detection: Use the POST or GET /v3/detect_faces methods to detect faces in images and get information about them, such as where the face is located in the image and the estimated age range and gender for each face. The service can also identify many celebrities by name and can provide a knowledge graph so that you can perform interesting aggregations into higher-level concepts.
  • Multi-faceted custom classifiers: You can now create and train highly specialized classifiers that are defined by several classes. For example, you can create a "new_red_car" classifier that is defined by the classes "new_cars" and "red_cars". To learn more about creating multi-faceted classifiers, see Structure of the training data.
  • Asynchronous training: Training of custom classifiers is now asynchronous, so training calls complete quickly while your custom classifier continues to learn in the background. To check on the training status of your custom classifier and find out when it is available for use, call the GET /v3/classifiers/{classifier_id} method and check the status response parameter.

2 December 2015

{: #2december2015}

The newest release of the {{site.data.keyword.visualrecognitionshort}} service is a breaking change that introduces a new version of the {{site.data.keyword.visualrecognitionshort}} API (v2). Version 1 of the {{site.data.keyword.visualrecognitionshort}} service is available for use until the service exits beta.

To immediately start using version 2 of the API, understand and update your code to reflect these changes:

  • In version 1 of the service, you analyzed images by labels. In version 2 of the service, labels are called classifiers. Classifiers have a unique classifier ID indicated by the classifier_id parameter and a short name indicated by the name parameter.
  • The POST /v1/recognize method for analyzing an image is now the POST /v2/classify method. The labels_to_check parameter is renamed to classifier_ids.
  • The GET /v1/tag/labels method for retrieving a list of labels in V1 is now the GET /v2/classifiers method for retrieving a list of classifiers.
  • In addition to retrieving a list of classifiers, you can also retrieve details for a specific classifier with the new GET /v2/classifiers/{classifier_id} method.
  • Version 2 of the Beta {{site.data.keyword.visualrecognitionshort}} API enables you to create custom classifiers with the new POST /v2/classifiers method. To learn more about creating custom classifiers, see Creating custom classifiers.
  • Version 2 of the Beta {{site.data.keyword.visualrecognitionshort}} API also enables you to delete custom classifiers with the new DELETE /v2/classifiers method.
  • Version 2 of the Beta {{site.data.keyword.visualrecognitionshort}} API requires the version parameter. Specify the release date of the version of the API you want to use in MM-DD-YYYY format.