Consider changing argument to a lowercase k? Python function and argument naming convention states that both should generally be lowercase and underscore separated.

Similarly, but outside the scope of just this function... consistency within a project is beneficial and there are already a small handful of functions that take a directory of things as an input. In the case of pcv.io.read_dataset the argument is named source_path, but in the case of pcv.transform.checkerboard_calib the argument is named img_path. In two other plantcv functions, pcv.transform.merge_images, pcv.visualize.pixel_scatter_plot, pcv.visualize.time_lapse_video, and pcv.segment_image_series uses slightly different logic where the arguments is paths_to_imgs , paths_to_imgs, img_list , and imgs_paths respectively and in all these cases, the user is expected to provide a list of filepaths rather than putting all input images into a single directory.

I personally think it's more user friendly to have the function take a directory rather than using pcv.io.read_dataset prior to running functions that take more than one image as input. I also think img_dir is succinct while descriptive but definitely open to discussion. @nfahlgren @maliagehan

Definitely can change to lowercase k, that was an oversight on my part.

Regarding the input format, I agree that consistency (especially in naming) is good, and that providing a path to a directory is more user friendly. However, I will just mention that the reason for changing to a list of image paths for pcv.transform.merge_images was because that function was written for the minirhizotron images and since there might be hundreds of tubes with 2-7 images per tube, it seemed cumbersome to require users to first put each set of images into a separate directory. Maybe there's a way to include a "prefix" argument similarly to the one in train_kmeans so that you can provide a directory to all images and then only include ones that contain a specified string? I don't know if that's necessarily more user-friendly.

Since several functions accept some type of directory or list of paths, we might want to spend some time working on consistency in a later PR.

The rationale for pcv.io.read_dataset is that we have one consistent way to get filenames from a directory dataset, rather than duplicating the code in many different functions, with or without different features (e.g. filtering, sorting, etc.). pcv.io.read_dataset currently allows filtering on a pattern and sorting (or not). Because it returns a list of file paths, one can do additional custom filtering if they need to. Functions that accept a directory are limited to what they are programmed to do and are not as flexible if the user needs something else, forcing them to rearrange the data instead.