Talk:CNNs in Image Segmentation

I think you did well in picking a pretty interesting topic. Your page seemed quite clearly written and your images were well chosen. The introduction was well written and you did a good job motivating the topic. It seems clear that you probably intend to expand on this page a bit more which will definitely help, as the descriptions of the papers are currently a bit too short for me to really get a handle on what's happening. It would also help to expand on the region proposal network section as well, or at least link to some further reading on the section maybe. There are also a few minor grammar/editing fixes. For instance, I think the images for Fig 1 and 2 should be flipped, and the references in the text to "authors of paper 1" should probably be filled in with the actual author names and paper name.

On a scale of 1 to 5, where 1 means "strongly disagree" and 5 means "strongly agree" please rate and comment on the following:

 The topic is relevant for the course. 5
 The writing is clear and the English is good. 5
 The page is written at an appropriate level for CPSC 522 students (where the students have diverse backgrounds). 4
 The formalism (definitions, mathematics) was well chosen to make the page easier to understand. 5
 The abstract is a concise and clear summary. N/A (not finished)
 There were appropriate (original) examples that helped make the topic clear. 2 (Could hopefully expand on things like the region proposal network section with some examples)
 There was appropriate use of (pseudo-) code. N/A
 It had a good coverage of representations, semantics, inference and learning (as appropriate for the topic). 4
 It is correct. 5
 It was neither too short nor too long for the topic. 2 (but it seems obvious you're intending to expand it)
 It was an appropriate unit for a page (it shouldn't be split into different topics or merged with another page). 5
 It links to appropriate other pages in the wiki. 4
 The references and links to external pages are well chosen. 3
 I would recommend this page to someone who wanted to find out about the topic. 4
 This page should be highlighted as an exemplary page for others to emulate. 3

If I was grading it out of 20, I would give it: 15

Hi Borna,

Nice writeup on Image segmentation using CNN. I have a few suggestions that I hope will help you. You may consider the following changes:

• Fill in Build’s On and related pages or remove those sections.
• Fig 1 and fig 2 seem to be interchanged
• May be providing a link to Dice’s score might be helpful. What is the significance of it and the ratio that you have specified? What is the intuition behind it? How does the interpretation change with Dscore in fig 3?
• How does segmentation identify an object when they are not labelled? Or how is a highest scoring region identified?
• How is a RoI pool layer different than a ROI Align layer. Can you write about that as you are comparing fig 4.a and fig 4.b What does the Mask branch do in fig 4.b?
• Overview of Paper 1 says that ROI Align and Parallel Mask are discussed before. But they are not. Maybe you should add about it.
• Why is the accuracy of a pixel classification necessary in image segmentation? What happens if there is an error?
• What is FCIS? The full form is not specified in the text - but as the fig 5 title. So one needs to search for it.
• How are the proposed regions of interest identified?
• One suggestion could be to write paper 2 as paper 1. This will let you identify the problem as “More errors were found in the results of the work in paper 1 because of …. They were solved in this paper 2 by ….” This is the incremental work of paper 2 over paper 1.

Cheers,