Talk:Class activation mapping

Reviewer 2

Hi, overall great article, well written and very complete. I just modified small details regarding typos, punctuation and other expressions.

I only have two main concerns:

- the introduction is very long, I believe it is not necessary to write such a long introduction. In particular, there are a couple of sentences for every concept such as DL, ML, CNNs and so on. These could be deleted as there are already specific articles on these topics, just link these concepts to the right page. Alternatively, they can go to the background section. More focus should be put on the actual topic of the article

- the figure in "Class model visualization and saliency maps for convolutional neural networks" seems misleading. Why is the saliency map highlighting the shadow of Di Caprio? It should highlight the most relevant pixels and, hence, the actor himself, not the background.

With these two modifications I think the article is ready. Natblida99 (talk) 13:33, 11 July 2025 (UTC)[reply]

Hi!

Very useful observations.

- You are right about the introduction. I wanted to give the overall view of where CAM methods are applied in the context of AI, without referring to other pages. When I was little I remember going back and forth an different pages, whenever something was not clear to me. I wanted to avoid this, but you are absolutely right. It's dispersive. I'll fix it.

- Again, you are right. I wrote a code with wrong coloration. I will update it as soon as I have the time to do it. I will delete the image for now.

Thanks a lot! TomMarine (talk) 08:13, 14 July 2025 (UTC)[reply]

Review 1

Quite a comprehensive article. Maybe even too much: the background on machine learning, the second leading section paragraph in particular, is redundant w.r.t. deep learning (and in part machine learning too) and could be significantly shortened by just relying on the hyperlink. I don't think there is a reason to re-explain to such lenght concepts already present elsewhere, a brief summary would be a better fit to effectively contextualize the article's main subject. Moreover, since you already have a "background" section, consider moving there such content, if you believe it to be strictly necessary (even tho I doubt it is, as for the above redundancy argument).

Too many acronyms, try to limit them to those that are wiledy recognized or that you use really often. E.g. keep CAM, GAP, XAI and CNN, while there is no need for DL, ML, AI, CV, VQA, that only occur in the leading sectin.

The ensuing explanation is extremely technical, I doubt it could be understood by a general audience. As such, I suggest you insert an introductory paragraph in the leading section or under "background" to clarify some gergon that you are gonna use and does not currently have dedicated wikipedia pages.

Consider adding the hyperlink: global average pooling.

Under both "Advantages and drawbacks" sections, it is not clear which reference is associated with which paragraph. Consider juxtaposing them properly, as to clarify <who said what>.

Missing "See also" and "Further readings" sections.

I hope to have been of help! EMJzero (talk) 13:36, 11 July 2025 (UTC)[reply]

Hi, thanks for the review!

I agree on the introductory part as too long, I will shorten it properly. Still I think that a little bit of contextual background may be useful, to avoid to rely too much on hyperlinks. I would define it as a personal preference. Avoiding open secondary pages is less noisy in my opinion. But still yes, too unnecesarily long.

I removed the acronyms you suggested.

For the technical explanation I don't entirely agree with you. I defined the explanations to present the high-level concept at the beginning, so that a general audience would grasp the main concept, and after that I defined the math description, to go deeper in the explanation. If one is not interested in the math explanation can skip the subparagraph and rely on the high-level concept described at the beginning.

For the references under "Advantages and drawbacks" sections, they all present the same advantages and drawback. Since I made a very general description I thought it would be easier to consider all the references with the same degree of importance. But still I agree on this.

Thanks for the analysis, it helped me a lot in strengthening my article! TomMarine (talk) 08:25, 14 July 2025 (UTC)[reply]

Reviewer 3

Hello,

Congratulations on this well-researched and detailed article. I must say it covers the topic thoroughly. However, I agree with reviewers 1 and 2 that the article is quite long, and this makes it challenging to follow.

As a researcher from a different field, I found it difficult to understand some of the content due to its technical depth and length. Simplifying the language where possible and reducing the overall length might make the article more accessible to a broader audience.

Hope that helps! Chauhanadi (talk) 14:26, 11 July 2025 (UTC)[reply]