If you display an image in your paper or distribute it in some way (like a dump of training data) you need permission to do so from the rights holder. If there's no rights because it's out of copyright, nothing needs to be done. For other content, it may be released under a license that lets you do those things without permission, or you may need to obtain permission individually, which may require payment. The license/agreement will likely also specify if crediting the owner of the image in some way is required.
Regardless of the terms governing image distribution, even if you just use an image without distributing it, you at least need to describe in your paper where you got your materials from. If a paper doesn't do that I'd call it a mistake. If it's peer reviewed then it's a mistake the reviewers missed, otherwise it's just a mistake by the authors. The general standard is that a paper should include enough information to replicate the results.