Altair RISE

A program to recognize and reward our most engaged community members

Nominate Yourself Now!

Confusion Using PyCustomBlock with ReadImage/ReadVideo and ShowImage in Twin Activate

Hi everyone, I'm exploring PyCustomBlock in Altair Twin Activate and running into some confusion around data flow and input/output port parameters.

Here's what I'm trying to do:

Use ReadImage or ReadVideo as the input source
Process the image/video in a Python block (PyCustomBlock)
Display the result using ShowImage

However, I’m stuck trying to understand:

Input Port Parameters: What row/column sizes should I configure in the Python block's element parameters to handle image/video data correctly?
Data Format Inside Python Script: Once the data enters the script, is it a serialized stream, a 2D NumPy array, or something else entirely?

To make things more confusing, I saw an example model in the Demo Explorer using rand(4,5) as input to PyCustomBlock, but its input port parameters are set to -1 for rows and -2 for columns. What do these values signify, and how do they affect data interpretation inside the script?

I’d greatly appreciate it if someone could clarify the recommended setup or point me to documentation that explains the data format flowing between blocks and how the port parameters tie into it.

Thanks in advance!

Find more posts tagged with

Twin Activate

Python

Accepted answers

All comments

RSGarciarivas

Hi Victor. There are a lot of blocks from the computer vision library that might be easier to implement than a python custom block. How exactly do you want to process images/videos with the custom block?

Regardless, I'll discuss with some colleagues and let you know how to properly handle this.

Rafael

VictorBravo9er

I want to do some custom processing on the pixel matrix data itself to calculate a threshold which would them be used to skew the original image.

PS: I would be performing NumPy matrix operations on that data, or even OpenCV!!!

VictorBravo9er

I did however figure out, using -ve sizes, to assign dynamic runtime resolved row/column sizes. But the problem is that the data being read isn't what I wanted. The PyCustomBlock reads a 1 x 1 matrix with a uint64 data, which after careful observation, I deduce, is just the timestamp of image read. However, the data I actually wanted is the actual pixel data that is being read.

If someone could help resolve this, please do🙏!!!!

RSGarciarivas

Hi Victor,

I ran some experiments and faced the exact same issue. Dealing with images in Python custom blocks is currently not supported because of this.

Alternatives include handling your processing purely with computer vision blocks or switching over to Altair AI Studio

Rafael

VictorBravo9er

Computer Vision blocks dont really have the operations that I want to execute. I want to process the data using numpy and opencv. Might even make use of tensorflow.

But, as you suggest, I'll give AI Studio a shot.

Thanks for your time and advice…

😅🤞