Course talk:CPSC522/StackedGAN

Comments

Based on the first impression, this is a good concise page, which seems to cover the essentials of both of the papers. What I am missing a bit is a differentiation of what is background and what is novel. I am assuming Geometry and style (should probably be "Style") is the new combination. The description seems a bit short for a paper which is over 20 pages long. Maybe adding a motivation and experiment results could add more substance? Secondly StackGAN seems to be presented rather matter-as-factish. There is little discussion on its concrete improvements over the preceding paper. Also it could have been interesting to discuss shortcomings for both approaches. The image in the two level GAN section is kind of hard to read. It may be beneficial to describe each individual stage and explain why it was chosen. In general, I feel like there is more to talk about the learning and experiments of the GAN^2 network. There are only few formulas, but that may be due to the fact that both papers are less theoretic.

Minor issues:

• Generative adversarial networks (GANs) "is" -> Generative adversarial networks (GANs) "are" ?
• ReLU and tanH should be linked to corresponding wiki articles.
• Kullback-Leibler should be explained or at least linked.

Marking Scheme

I 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. 4 (sometimes a bit colloquial "While there is a page on GANs within this wiki, this is also a good explanation of how they work.")
   The page is written at an appropriate level for CPSC 522 students (where the students have diverse backgrounds). 5 (Using the GAN page as background, yes)
   The formalism (definitions, mathematics) was well chosen to make the page easier to understand. 3 (there is little formalism)
   The abstract is a concise and clear summary. 5
   There were appropriate (original) examples that helped make the topic clear. 5 (I like the birds, there is not much more to add)
   There was appropriate use of (pseudo-) code. 5 (Does not require code, so that is fine)
   It had a good coverage of representations, semantics, inference and learning (as appropriate for the topic). 4 ( I thought it was sufficient)
   It is correct. 5 (As far as I can tell, yes)
   It was neither too short nor too long for the topic. 3 (A bit short, could be expanded)
   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. (5)
   The references and links to external pages are well chosen. (5)
   I would recommend this page to someone who wanted to find out about the topic. (4 It provides a good overview, but leaves me a bit confused.)
   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: 16

The page is informative, concise and does a good job of covering stackGAN for targeted generation. Some of my suggestions to improve the page:

1. A little elaboration of the KL-divergence regularization might be required
2. Some more results from the papers would have been nice
3. Mathematical formalism is lacking, though that's because you assume prior knowledge of GAN.

• 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). 5
• The formalism (definitions, mathematics) was well chosen to make the page easier to understand. 4
• The abstract is a concise and clear summary. 5
• There were appropriate (original) examples that helped make the topic clear. 4
• There was appropriate use of (pseudo-) code. 4
• It had a good coverage of representations, semantics, inference, and learning (as appropriate for the topic). 5
• It is correct. 5
• It was neither too short nor too long for the topic. 4
• 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. 5
• The references and links to external pages are well chosen. 5
• I would recommend this page to someone who wanted to find out about the topic. 5
• This page should be highlighted as an exemplary page for others to emulate. 4

If I was grading it out of 20, I would give it: 18