v3.14.7

What's Changed

• [fix] Auto task abilities: allow client to set the mode to image or video by @tschulz in #175

Full Changelog: v3.14.6...v3.14.7

v3.14.6

What's Changed

• [VREE-219] Simplified auto_task replacing the deprecated auto_prompt field when creating an ability group by @tschulz in #174

Full Changelog: v3.14.4...v3.14.6

v3.14.4

What's Changed

• [feature] Allow fps overwrite for every source (upload/load_from etc) by @tschulz in #173
• Fix type VlmAbility and return type for resolveAliases by @zairavm in #172

Full Changelog: v3.14.3...v3.14.4

v3.14.3

Overview

Introduce named parameters in WorkerEndpoint.process()

To support future optional parameters, the process method using positional parameters WorkerEndpoint.process(source: Source, params?: ComponentParams[]) has been deprecated in favor of WorkerEndpoint.process({ source, componentParams, roi }: ProcessParams).

Support roi parameter in endpoint.process()

Using the newly introduced named parameter version of process() a client can pass a rectangle ROI (in image pixel coordinates) to limit inference onto that area. The resulting predictions will continue to be in the original full image pixel coordinate system. Just the attention is masked and will only return predictions in the ROI area.

Support motion detection

New named option motionDetect in WorkerEndpoint.process() to control motion detection for videos and live streams.

• motionDetect=true to switch on motion detection
• motionSensitivity sensitivity of motion detection per cell, 0.0-1.0 default 0.5
• motionThreshold threshold percentage of cells with motion to trigger a motion event, 0.0-1.0 default 0.01
• motionGap seconds w/o motion before motion ends, default 1
• motionGridX X dimension grid size, default 10
• motionGridY Y dimension grid size, default 10

What's Changed

• [OPA-28] Support roi parameter in endpoint.process() by @tschulz in #170
• [OPA-27] Motion detection based on source parameters @tschulz in #171

Full Changelog: v3.14.1...v3.14.3

v3.14.1

What's Changed

• [VREE-200] Add DataEndpoint.resolveAliases() method to support cloning public abilities by their alias by @tschulz in #169
  • DataEndpoint.resolveAliases(aliases: string[]) returns an equal length array of AliasResolution. Each element of the response array has one of the requested alias values and the resolved model_uuid if found. if the alias is not found, model_uuid will be null. The order of the return array is not guaranteed.

Full Changelog: 3.14.0...v3.14.1

3.14.0

What's Changed

Vlm Abilities Auto-Prompt support (create/refine)

Creating a new Vlm Ability Group now supports and a new attribute auto_prompt. The default behavior of DataEndpoint.createAbilityGroup() has not changed. When give a valid auto_prompt attribute, the call will auto
create an ability as child of the new ability group and start a background job to analyze the denoted dataset and
based on the ground-truth of the images and/or videos create a first version of ability config.

A new state in_progress is introduced which will be the abilities state until the auto-prompt process completes.
At that time the ability state will change to draft as if the ability was fully created by the client.

Example CREATE ability via auto-prompt:

await endpoint.createVlmAbilityGroup({
    name: ' ... ',
    description: ' ... ',
    auto_prompt: {
        num_samples: 5,
        infer: {
             'worker_release': 'qwen3_instruct',
             'transform_into': {
                  classes: ['dark', 'light']
             }
        },
        evaluate: {
            dataset_uuid: ' ... '            
        }
    }
})

Refining an existing ability via auto-prompt is a new feature that can be invoked on any ability, whether it was created
via explicit (manual) configuration or via auto-prompt. Example REFINE ability via auto-prompt:

await endpoint.refineVlmAbility(
    ' ... ability uuid .... ', 
   {
        num_samples: 5,
        infer: {
             'worker_release': 'qwen3_instruct',
             'transform_into': {
                  classes: ['dark', 'light']
             }
        },
        evaluate: {
            dataset_uuid: ' ... '            
        }
    }
)

Vlm Abilities Clone

Exisiting Vlm Abilities can be cloned into a newly created ability group. The caller must have read access to the
source ability. Either via a shared account or if the ability is marked as public like abilities in the Ability Hub.

The call DataEndpoint.cloneVlmAbility() inherits all attributes of the source ability and allows the caller to optionally
overwrite name and to set a default_alias_name.

Other behavior change of Vlm Ability endpoints:

• DataEndpoint.publishVlmAbility() will always assing the next semver version and the latest tag if no tag name is given.
• All resonse objects of type VlmAbility have a new attribute auto_prompt if the ability was created or refined by auto-prompt, for all other abilities this attribute is undefined
• VlmAbilityState enum has a new value in_progress

Full Changelog: v3.13.3...3.14.0

v3.13.3

What's Changed

• VREE-179 listAssets add filtering options by @zairavm in #166

Full Changelog: v3.12.2...v3.13.3

v3.12.2

What's Changed

• [BUG-203] Wait server side for a new compute session to be ready to avoid excessive polling. by @tschulz in #165

Full Changelog: v3.12.1...v3.12.2

v3.12.1

What's Changed

• [VREE-183] Asset.rois support contour type areas in addition to rectangle types by @tschulz in #162

Full Changelog: v3.12.0...v3.12.1

v3.12.0

What's Changed

• [VREE-186] Add support for Rois on Assets by @tschulz in #161

Full Changelog: v3.11.0...v3.12.0