OpenAdaptAI / OpenAdapt

Open Source Generative Process Automation (i.e. Generative RPA). AI-First Process Automation with Large ([Language (LLMs) / Action (LAMs) / Multimodal (LMMs)] / Visual Language (VLMs)) Models
https://www.OpenAdapt.AI
MIT License
880 stars 115 forks source link

feat: Enhance take_screenshot for multi-monitor support #792

Open onyedikachi-david opened 3 months ago

onyedikachi-david commented 3 months ago

Fixes #766 /claim #766

What kind of change does this PR introduce?

Feature

Summary

This PR introduces support for multi-monitor setups in the take_screenshot function. It includes the following changes:

Checklist

How can your code be run and tested?

To run and test the code:

  1. Ensure you have the necessary dependencies installed.
  2. Run the test suite using pytest:
    pytest tests/openadapt/test_monitors.py
  3. Verify that all tests pass, including the new tests for multi-monitor support.

Other information

No additional context needed.

abrichr commented 3 months ago

Thank you @onyedikachi-david ! What do you think about saving taking a screenshot of all of the available monitors, rather than just the active one? This can provide useful additional context to the model if something changes on one of the screens.

onyedikachi-david commented 3 months ago

Thank you @onyedikachi-david ! What do you think about saving taking a screenshot of all of the available monitors, rather than just the active one? This can provide useful additional context to the model if something changes on one of the screens.

Okay, I'll implement it. just that I was being mindful of how it is being used across the codebase. If there are more than one monitors, then a tuple or list will be returned, which will affect the screenshot function usage across the codebase.

abrichr commented 3 months ago

Thank you please do. Maybe keep the current functionality behind a config param.

onyedikachi-david commented 3 months ago

Thank you please do. Maybe keep the current functionality behind a config param.

New take_screenshot function using config:

def take_screenshot() -> list:
    """Take screenshots of the current monitor or all available monitors.

    Returns:
        list of PIL.Image.Image: A list of screenshot images.
    """
    with mss.mss() as sct:
        monitors = sct.monitors
        screenshots = []

        if config.CAPTURE_ALL_MONITORS:
            for monitor in monitors[1:]:  # Skip the first entry which is a union of all monitors
                sct_img = sct.grab(monitor)
                image = Image.frombytes("RGB", sct_img.size, sct_img.bgra, "raw", "BGRX")
                screenshots.append(image)
        else:
            current_monitor = get_current_monitor(monitors)
            sct_img = sct.grab(current_monitor)
            image = Image.frombytes("RGB", sct_img.size, sct_img.bgra, "raw", "BGRX")
            screenshots.append(image)

        return screenshots

If I got you right, I should go ahead and change it to have a list return type, if that is the case, then its current implementation in models.py

 @classmethod
    def take_screenshot(cls: "Screenshot") -> "Screenshot":
        """Capture a screenshot."""
        image = utils.take_screenshot()
        screenshot = Screenshot(image=image)
        return screenshot

will change to this:

@classmethod
    def take_screenshot(cls: "Screenshot") -> list:
        """Capture a screenshot."""
        images = take_screenshot(all_monitors=all_monitors)
        screenshots = [cls(image=image) for image in images]
        return screenshots

And its usage will also change:

screenshots = Screenshot.take_screenshot(all_monitors=True)
for idx, screenshot in enumerate(screenshots):
    screenshot.image.show(title=f"Monitor {idx + 1}")
abrichr commented 3 months ago

I think we want to return a single PIL.Image containing all screenshots. Any way to determine relative positioning of the screens?

onyedikachi-david commented 3 months ago

I think we want to return a single PIL.Image containing all screenshots. Any way to determine relative positioning of the screens?

Yes, it is best to return PIL.Image. Let me think of a way to go about it.

onyedikachi-david commented 3 months ago

@abrichr I just implemented the requested changes.

onyedikachi-david commented 3 months ago

@abrichr I don't know if you've found time to review the PR.

onyedikachi-david commented 2 months ago

Thank you for putting this together @onyedikachi-david ! And thank you for your patience.

I left a few comments concerning performance. Because we call take_screenshot in a tight loop during recording, it's important that it is as performant as possible.

Can you please run python -m openadapt.record with CAPTURE_ALL_MONITORS = True and = False (while recording similar behavior, e.g. opening the calculator and clicking 2 x 3), and paste the performance plots here? (The path to the performance plot is logged to stdout at the end of the recording.)

You're welcome. Thanks for the feedback and all. Two issues though:

  1. I don't have an external monitor to test.
  2. This one isn't much of an issue though, I'll see how to run it on my Linux machine (although it didn't work last time I tried); my Mac OS crashed yesterday.

Just give me awhile to implement the requested changes.

onyedikachi-david commented 2 months ago

Please review @abrichr, just implemented the requested changes, no screenshot yet though. Yet to fix out my Mac OS crash

onyedikachi-david commented 2 months ago

@abrichr I'm sorry for keeping a stale PR. I hope it isn't blocking anything. I'm still trying to get my macOS setup together. I was using a Hackintosh (dual-booted with Linux), but it got corrupted while I was reinstalling Docker. I'm still figuring out how to resolve this without making things worse. I don't want to lose my Linux files/partition, which is still functional.

abrichr commented 2 months ago

@onyedikachi-david thank you for the update. For now it is not blocking. 🙏

onyedikachi-david commented 2 months ago

@onyedikachi-david thank you for the update. For now it is not blocking. 🙏

Okay.