YOLOX | performance anchor-free YOLO | Computer Vision library

It appears that the 'visions.application' module was available in v0.7.1

https://github.com/dylan-profiler/visions/tree/v0.7.1/src/visions

But it's no longer available in v0.7.2

https://github.com/dylan-profiler/visions/tree/v0.7.2/src/visions

It also appears that the pandas_profiling project has been updated, the file summary.py no longer tries to do this import.

In summary: use visions version v0.7.1 or upgrade pandas_profiling.

It seems that there's already an open discussion with the Google team to get this Feature Request addressed:

https://issuetracker.google.com/154156890

I would recommend you to comment on the Public issue tracker and indicate that "you are affected to this issue" to gain visibility and push for get this change done.

Other that that I'm unsure is that can be implemented locally.

Finally I could do it. I had to set chessboard_size = (12,7) then it worked. I had to count the internal number of horizontal and vertical corners.

If you could make all pixels outside of the contour transparent then you could use CIKmeans filter with inputCount equal 1 and the inputExtent set to the extent of the frame to get the average color of the area inside the contour (the output of the filter will contain 1-pixel image and the color of the pixel is what you are looking for).

Now, to make all pixels transparent outside of the contour, you could do something like this:

1. Create a mask image but setting all pixels inside the contour white and black outside (set background to black and fill the path with white).
2. Use CIBlendWithMask filter where:
   • inputBackgroundImage is a fully transparent (clear) image
   • inputImage is the original frame
   • inputMaskImage is the mask you created above

The output of that filter will give you the image with all pixels outside the contour fully transparent. And now you can use the CIKMeans filter with it as described at the beginning.

BTW, if you want to play with every single of the 230 filters out there check this app out: https://apps.apple.com/us/app/filter-magic/id1594986951

CIFilters can only work with CIImages. So the mask image has to be a CIImage as well. One way to do that is to create a CGImage from CAShapeLayer containing the mask and then create CIImage out of it. Here is how the code could look like:

This is my complete code:

For some reason, remove

As they said in WWDC 2019 video, Text Recognition in Vision Framework:

First, a prerequisite, you need to check the languages that are supported by language-based correction...

Look at supportedRecognitionLanguages for VNRecognizeTextRequestRevision2 for “accurate” recognition, and it would appear that the supported languages are:

You are on the right track, but I have to be honest. Without DeepLearning you will get good results but not perfect.

That's what I managed to get using contours:

Code:

For LabVIEW users who have the NI vision library installed, there are VIs that allow for the image data of an IMAQ image to be copied from a 2D array.

For single-channel images (U8, U16, I16, float) the VI is

Vision and Motion >> Vision Utilites >> Pixel Manipulation >> IMAQ ArrayToImage.vi

For multichannel images (RGB etc) the VI is

Vision and Motion >> Vision Utilites >> Color Utilities >> IMAQ ArrayColorToImage.vi

Example 1

An example of using the IMAQ ArrayToImage.vi is shown in the snippet below where U16 data is read from a binary file and written to a Greyscale U16 type IMAQ image. Please note, if the file has been created by other software than LabVIEW then it is likely that it will have to be read in little-endian format which is specified for the Read From Binary File.vi

Example 2

A similar process can be used when some driver DLL call is used to get the image data as a buffer. For example, if the driver has a function capture(unsigned short * buffer) then the following technique could be employed where a correctly sized array is initialized before the function call using the initialize array primitive.