Oakutils

Added

- blobs.testing.BlobEvaluator introduced
- Exposed more compiler functionality
- blobs.get_model_path
- blobs.get_cache_dir
- blob.get_model_name
- Added blobs.models.bulk for easier bulk importing of model paths
- i.e. bulk.GAUSSIAN_15x15 is a tuple of all compiled models
of that function
- ALL_MODELS, as a list of each valid tuple of paths
- Added new script `model_visualizer` which allows visualizing a variety
of pre-compiled models.

Improved

- Explicit call to VPU.stop in blobs.benchmark_blob.
- Enabled with statement usage with VPU to prevent missing stop calls.
- Rewrote all existing tests to use new BlobEvaluator
- More debug logging in VPU and node creation

Added

- blobs.benchmark_blob
- Measure the latency of a compiled blob.
- nodes.get_yolo_data
- Create a YolomodelData object from a json file.

Added

- benchmark submodule
- measure_latency and measure_throughput functions.
Allow measuring roundtrip latency and max
uplink and downlink throughput.
- Info functions for compiled blobs
- Can get the layer information for input/output layers
of compiled blob files.
- Added function for getting blob objects via paths
with verification.

Improvements

- create_device
- Now accepts a usb speed parameter.

Added

- Supported for OAK-1 devices through the calibration API
- get_camera_calibration now reads device type and pulls
correct information.
- ApiCamera, LegacyCamera, VPU, and Webcam now accept
a device_id for specifying which physical OAK device
to utilize.
- create_device: A function for creating a DepthAI Device
from a pre-defined pipeline. Accepts parameters for
where to create said device.
- nodes.buffer submodule
- Buffer, MultiBuffer, SimpleBuffer classes. Each class
enables methods for making DataInput/OutputQueues easier
to use in bulk.
- create_synced_buffer, allows wrapping multiple output
streams with a single function getting a single output
from each stream at once.

Improvements

- VPU methods refactored to handle multiple neural networks
- reconfigure parameters changed
- Added reconfigure_multi
- Function signature changed for run
- Pipeline generation and VPU thread use new backend

Changed

- get_camera_calibration
- New return type of CalibrationData | ColorCalibrationData
This comes from the OAK-1 support
- Could break old code which utilized this function, but
would primarily affect static type-checkers like mypy
- Can replace with get_oakd_calibration to achieve 1:1

Changed

- Removed Open3d as a strict requirement
- point_clouds submodule only imported if open3d is found
- Pinhole models in calibration will be None unless open3d is found
- If open3d is found then oakutils will function identically as pre 1.5
- open3d can be installed with oakutils[o3d]
- Added cv2ext as a strict requirment
- Switched the DisplayManager to use cv2ext.Display

Added

- get_nn_data function
- allows getting generic data from an executed neuralnetwork

Fixed

- VPU
- Incorrect return type for run
- blob_path listed incorrectly as parameter for init
- Changed reconfigure to accept str | Path
- Documentation
- Resolved indentation errors in docs

Changed

- Removed README from doc homepage