Description

This PR implements a new version of the camera focus block to provide improved focus measures across a wider range of images and use cases, including help adjusting exposure and aperture (focus metrics are also impacted by under/over-exposed images).

This PR contains a V2 workflow block that:

1. Implements the the Tenengrad focus measure that is less sensitive to noise than the existing Brenner measure and provides a monotonically increasing value that is more useful for sorting + ranking images than Laplacian variance
2. Accepts an optional list of detections to return the focus measure in specific bounding boxes, allowing users to measure the focus of objects of interest
3. Displays an exposure histogram to the visualization to help differentiate between inadequate aperture/exposure settings and focusing that may be negatively impacting the sharpness of the image
4. Provides an optional zebra stripe visualization to highlight overexposed and underexposed areas more prominently and some small conveniences like a grid and crosshairs for making sure the camera is aligned

Since this is a pretty significant change in behavior, this is a V2 with a new implementation.

Type of change

Please delete options that are not relevant.

• New feature (non-breaking change which adds functionality)
• This change requires a documentation update

How has this change been tested, please provide a testcase or example of how you tested the change?

Tested this locally using a USB webcam below 👇

Tested using bounding box and label visualizations containing the focus measure of the boxes:

Any specific deployment considerations

Since this is a classical CV block that uses a standard OpenCV Sobel operator on CPU, it can be deployed anywhere inference can run.

Docs

Currently, all documentation for this block is generated from code - I will be writing a followup blog post as part of Shipmas to update the information from this previous post: https://blog.roboflow.com/computer-vision-camera-focus/