Allow processing to continue if an error occurs in image decoding

[jellehierck]

When an error arises during image decoding, processing stops entirely. It would be nice to have an option to only report the error but continue processing other messages.

Concretely, we have rosbags with many images in them. Unfortunately, some of the images are an empty sensor_msgs/msg/Image. An empty image does not have any data, no size, and also no encoding field. That triggers the error in ros2 unbag.

However, since ros2 unbag is multi threaded, it will stop other threads too when it encounters the error, which causes a race condition and other images to only be partially saved (which explains the inconsistent gray areas in other saved images).

One workaround is to set the CPU usage in ros2 unbag to 0 %. This does not prevent the error, but all images before the empty image will be processed and saved correctly. However, the program will still stop processing subsequent images.

My other (very, very dirty) workaround is to slightly adjust

ros2_unbag/ros2_unbag/core/utils/image_utils.py

Line 79 in ae73136

raise ValueError(f"Unsupported encoding: {encoding}")

to prevent the error from being raised.

    # Around line 80
+   return np.zeros((2,2))
-   raise ValueError(f"Unsupported encoding: {encoding}")

[lostendo]

Hi @jellehierck,

I was initially hesitant to add this feature, as the ideal solution would be to address the root cause of the empty messages or prevent erroneous topics from being exported entirely.

However, I’ve implemented a workaround as a settable flag on a separate branch. We’ll evaluate whether this makes it into the main branch in the future.

[jellehierck]

Thank you for the very fast response and solution! I agree that defective input data is the root cause, but I'm happy with this addition so we can use your tool to inspect even defective data.

I'll play around with it next week and let you know if I encounter anything unusual.

[jellehierck]

After some testing, I find the addition of the continue on error flag very useful.

My initial confusion that led to the creation of this issue was that exported images seemed to be corrupted, because they were only partially exported (see example image below). I now realize that this was because of parallel workers which were interrupted because another worker threw an exception.

To clarify this for other users of ros2 unbag, I created #35 to make error messages during image export more descriptive.

[jellehierck]

Another issue when running the tool from the CLI: an error NameError: name 'logger' is not defined

The problem is in line https://github.com/ika-rwth-aachen/ros2_unbag/blame/075f06da2f8bc99718766c0902b5b24176dc30d7/ros2_unbag/export.py#L201

The undefined logger variable is also present multiple times around https://github.com/ika-rwth-aachen/ros2_unbag/blame/075f06da2f8bc99718766c0902b5b24176dc30d7/ros2_unbag/export.py#L736

Could that be replaced with a simple print() instead?

[lostendo]

The logger fix and your changes from #35 are now on the branch. I’m not sure yet if these will actually be merged into main.