[seb5g]

Before: the Update ROI from Viewer and Apply ROI led_push of CameraBase only reacted to a change of the ROI in the viewer: clicking Update did not take the ROI already drawn, and the Image width / Image height were not recomputed when a ROI was applied. The LEDs only said on / off.

After: both are action_led parameters (a button and a LED with the PyMoDAQ status colors): grey when not activated, blue while the ROI is applied, green when activated, orange on error (no ROI in the viewer, ROI outside the sensor, the camera refuses it).

• Update ROI from Viewer takes the ROI now in the viewer, and the ROI of the viewer is followed (green) until Clear ROI+Bin.
• Apply ROI applies it; once green, the ROI is applied again each time it changes in the viewer.
• Image width / Image height are updated with each ROI and binning.
• Also: color_conversion has a default value ('None'): with OpenCV installed and no default, the conversion looked for cv2.COLOR_None. This made 7 tests of test_camera_base.py fail where cv2 is installed (they failed on 5.0.x too).

How: CameraBase connects sigActivated of the two parameters, which needs PyMoDAQ/PyMoDAQ#1431 (the activation of a button reaches the plugin thread); until it is released the buttons do nothing in the application, and the pymodaq floor of this package has to be raised to the release that contains it. Tests (test_camera_base.py, activation done with Parameter.activate() on the plugin parameter): LED states in each case, the ROI taken from the viewer on click, the image size, the ROI outside the sensor, a camera that refuses the ROI, following the viewer until cleared. 53 passed with PyMoDAQ #1431 from source; test_plugin_package_structure fails locally because the package is not installed. Not tested with a real camera or a GUI click.

---

Generated by Claude Code

[seb5g]

All jobs pass except compatibility / windows-latest / Python 3.12 / pyqt6, which stopped printing after test_clear_roi_goes_back_to_the_full_sensor and was cancelled after about 30 minutes, with test_roi_change_during_live_restarts_it_with_the_new_roi as the last test running. That test existed before this PR and it passes here on Linux (6 runs in a row), so this looks like a hang of the live thread on the Windows runner rather than an assertion failure. I did not find a cause in this PR's diff, which only touches the ROI buttons and compute_axes. I re-run the cancelled job once. If it hangs again I will look at whether the live test needs a guard on Windows.

---

Generated by Claude Code