Course Analysis, DD2422 Image Processing and Computer Vision, 6 credits

Course given 2007/2008 in study period 3

  • Course responsible:
    Danica Kragic
  • Lecturer:
    Danica Kragic
  • TAs:
    Per Rosengren, Oscar Danielsson, Javier Romero
  • Lectures:
    total 16 lectures (32 hours), out of which 4 hours has been spent for scheduled exercises in the lecture rooms.
  • Laboratory work:
    In total, 72 hours of help was given during the laboratory work. Labs were performed in CSCs computer rooms.
  • Registered students:
    58 undergraduate students have registered for the course.
    55 followed the course (did at least one of the labs).
  • Course material:
    Book:
    "Digital Signal Processing", Gonzalez/Woods Prentice Hall (2008, 3rd edition)
    All lectures were accessible for download.
    All laboratory work has been posted online, both in English and Swedish version.
  • Examination and course requirements:
    The assessment was one exam and three labs. The course is 6p in total, out of which the exam is 3p and the labs are 3p.
    To meet the requirements of the new grading system, the final exam consisted of Part A and Part B. For the passing grade E, a student had to answer correctly at least 80% of Part A. If the score was less than 80%, the rest of the exam was not corrected.
    Part B of the exam consisted of seven exercises that gave at most 50 points.
    The bonus credits from the labs (maximum 5) were added to Part A if a student dis not reach 80% - otherwise they were added to Part B.
    Grading:
    >48 A
    >38 B
    >29 C
    >17 D

    Part A:
    >80% E
    >75% FX

    Allowed helping material: Calculator, the mathematics handbook Beta (or similar) and a handwritten (not scanned or coppied) A4 paper with notes. The answers were given either in English or Swedish.
    The exam was corrected 2 weeks after it was held. One person complained aboutr not being considered all the extra lab credits.
  • Course results:
    34 undergraduate students completed a part or all of the necessary examination parts by the end of the course.
  • "Prestationsgrad":
    73.6 % (Out of the students that followed the course)
    100% have fulfilled lab1
    89.1% have fulfilled lab2
    67.3% have fulfilled lab3

    49 students have taken the exam and the grades were:
    10.2% - A
    20.4% - B
    30.6% - C
    22.4% - D
    10.2% - E
    6.2% - F
  • "Examinationsgrad":
    63.6%
  • Relation to the previous years:
    Compared to the previous year, most of the lectures were updated. Some of the old material was removed and new examples have been included in the lectures.

    In particular, Fourier transform lectures were further shortened. Even with a number of descriptive examples, some of the students still feel that the lectures are too theoretical. Only a few students have previously taken some course that handled this type of transforms. Since Fourier transform is basis for the Lab 2, a more thorough gothrough will be needed next year. The lecture with Principle Component Analysis was also updated and more descriptive examples were given. It was also recognized by some of the students that this lecture was interesting and good.

    Lecture on multi-scale representations and feature extraction was completely rewritten.

    Similar to the previous year, all the material was available in both Swedish and English. This was appreciated by the students. It resulted with fewer questions and misunderstandings. Again, some students have chosen to work individually which has required some extra hours for 'redovisning'. However, having the material in both languages is easier for students and we will continue with it.

    No 'kursbunt' was printed. This was the third time since the course was initiated. For many years, students have complained about the costs which were high since they have to pay for both the book and the 'kursbunt'. Therefore, from this year all the material was available for download and some links to were the additional material can be found were give.

    Lab notes were updated based on the previous year comments.

  • Summary:
    16 lectures were held and most of them were followed by more than 80% of students. During the lectures, students were posing questions both in Swedish and English and this has been found as a good side in the student evaluation. Students were supposed to read the book on their own in parallel with lectures but more than 50% have said in the evaluation that they did not prepare for the lectures.

    Results of students evaluation:

    16 students filled in the evaluation form (29%).
  • Course in general:
    More than 82% thought that the course was interesting and fruitful.
  • Course level:
    Most students feel that the course level was medium or difficult. Students complain about statistics and Fourier transform - even if these lectures have undergone most changes compared to the previous years. The lectures very completely rewritten and more examples have been given.
  • Planned changes:
    One of the obvious things we will continuw with is to keep both Swedish and English webpages. This is also valid for all the laboratory work, the exam, all the lecture notes as well as the coursePM that were written in both languages. CoursePM was very detailed and it has included all the information that was written on the webpage.

    Some small changes were applied to laboratory notes. The notes include both the theoretical and experimental parts. The students are supposed to do the work in MATLAB and answer a number of questions. The notes were changed so that it was more clear that answers to the questions have to be provided in written text directly in the laboratory notes. It has been mentioned by some of the students that some of the questions have not been very clear. We keep a list of proposed changes, which will be introduced in the notes for the next year. A more extensive change will be to attach the expected, final results in the notes. These results may be both quantitative and qualitative (final processed images).

    The course is a 6 credit course. Judging from the student evaluation, from this and previous years, 7.5p or more would be suitable regarding the current material.



    Raw student evaluation of the course: Image Processing and Computer Vision

    DD2422 Image Processing and Computer Vision

    Results of students evaluation


      A few questions hoping for honest anonymous answers and comments (questions can be answered in Swedish, English, or Croatian).


    1. Was this course interesting and fruitful?

      1. 41% (7 st) Yes, very.
      2. 41% (7 st) Yes.
      3. 12% (2 st) Neutral.
      4. 0% (0 st) Not very.
      5. 0% (0 st) No.


    2. Do you regard the course as easy or difficult?

      1. 0% (0 st) Very easy.
      2. 18% (3 st) Easy.
      3. 29% (5 st) Ok.
      4. 41% (7 st) Difficult.
      5. 6% (1 st) Very difficult.


    3. Where the goals of the course clear to you early on in the course?

      1. 71% (12 st) Yes.
      2. 12% (2 st) Don't remember.
      3. 12% (2 st) No.


    4. The prerequisites of the course correspond to the compulsory courses in maths, computer science and numerical analysis. Was your previous knowledge enough to follow the course?

      1. 76% (13 st) Yes.
      2. 12% (2 st) Not sure.
      3. 6% (1 st) No.

      Additional comments on prerequisites:

      good knowledge of fourier transform and MatLab was needed too.
      ---
      Vissa saker var lite oklara. En del av deltagarna hade läst ex. maskinlärning, och då antogs alla ha gjort det.


    5. Did you use the course book 'Digital Image Processing' of Gonzalez and Woods?

      1. 59% (10 st) Yes, bought it.
      2. 24% (4 st) Yes, borrowed it.
      3. 12% (2 st) No, didn't use it.


    6. What do you think of 'Digital Image Processing' of Gonzalez and Woods?

      1. 35% (6 st) Very good.
      2. 35% (6 st) Good.
      3. 12% (2 st) Ok.
      4. 6% (1 st) Not so good.
      5. 0% (0 st) Poor.
      6. 6% (1 st) Haven't used it.

      Additional comments on the course book:

      expensive and nor fully covered in the class.
      ---
      Lite lång, som alla amerikanska böcker.
      ---
      No informations on the camera model.
      The maths are too simple.

      ---
      En del av de termer som det frågas om på (ex)tenta-frågorna finns inte med i sakregistret till boken. T.ex. "Optimal flow".
      ---
      It's a very verbose book. Personally I percieved it to have a bad signal to noise ratio. A book that is more to the point (maybe given as an alternative?) would be appreciated.
      ---
      Isn't there a book that covers more on computer vision? The lecture notes are not very structured and complete, so it would be useful to have a reference book for finding details.
      ---
      The book is simply perfect!


    7. What do you think about the lecture notes?

      1. 12% (2 st) Very good.
      2. 29% (5 st) Good.
      3. 35% (6 st) Ok.
      4. 18% (3 st) Not so good.
      5. 0% (0 st) Poor.
      6. 0% (0 st) Haven't used them.

      Additional comments on lecture notes:

      Skulle kunna varit numerade, om man skulle anteckna något viktigt som stod på en speciell sida.
      ---
      Some points are too developped (fourrier transform), and others are overlooked (SIFT features)
      ---
      Kändes lite faktafattiga ibland. Ett stort plus för att de gick att titta igenom innan föreläsningen så man kunde förbered sig.
      ---
      Some things are mentioned to briefly.
      ---
      Alot of information in a small space, especially annoying since alot of the topics were not in the book.
      Had to construe information from the not-so-good lecture notes and out of context handouts.

      ---
      The lecture notes should be more structured. It is unclear where the explanation of one concept ends and the next begins.
      ---
      English could have been less sophisticated in lecture notes
      ---
      It was very difficult to follow the material in the lecture notes without attending the lectures. This is true in general for the lectures after edge detection and for classification in particular. I do not know if this is intentional, but it is not a good thing. If I happen to miss a lecture, it is nice to be able to read through the lecture slides and understand (mostly) everything.


    8. How about the lab notes?

      1. 18% (3 st) Very good.
      2. 47% (8 st) Good.
      3. 12% (2 st) Ok.
      4. 18% (3 st) Not so good.
      5. 0% (0 st) Poor.
      6. 0% (0 st) Haven't used them.

      Additional comment on lab notes:

      i would like to have a bit more help on the 3rd lab.
      ---
      Frågorna i labbinstruktionerna var utydligt formulerade. Många handlade om att göra något trivialt och sedan dra slutsatser om detta. Skulle vara bättre med tydligare, mer tekniska frågor och färre filosofiska frågor.
      ---
      De svenska var konstigt översatta från engelska. En del gånger var det också svårt att veta vad man skulle göra.
      ---
      The labs are well constructed, but maybe they should be more problem oriented.
      ---
      There are a few typos here and there along with some ambiguities, but nothing major.
      ---
      Sometimes diffuse formulations. I'm aware of that you don't want to give away to much information sometimes and that's not what I mean.
      ---
      The difficulty of the labs was most often to decipher what was meant by the questions.
      ---
      Some elaboration is needed regarding the differences between filter2 and conv2 in lab3. I still have not understood it completely, despite having a discussion with an assistant.


    9. How many lectures did you attend?

      1. 0% (0 st) Less than 20%.
      2. 6% (1 st) 20-40%.
      3. 0% (0 st) 40-60%.
      4. 24% (4 st) 60-80%.
      5. 65% (11 st) More than 80%.


    10. What do you think of the lectures (the topic and they way the material is presented)?

      1. 24% (4 st) Very good.
      2. 35% (6 st) Good.
      3. 29% (5 st) Acceptable.
      4. 0% (0 st) Not so good.
      5. 6% (1 st) Poor.
      6. 0% (0 st) Haven't participated.

      Additional comments:

      I attended to the five first lectures, but I found them too long for their content.
      ---
      Although the lecture about stereo and essential matrices was a bit confusing
      ---
      In my view, this is the part of the course that could use the most improvement. You also seem to be a teacher who wants to improve, so I have written down some of the observation that I have made during class.


      Danica:

      * The slides often lack in structure. It was typically hard to tell whether we were listening to a continuation of the topic of the previous slides or not. You could spend the first few minutes introducing the main topics/methods of the lecture, in what order they will be coming (and the relationship between them if possible), and then make it clear when we're switching from one topic to the next.

      * Your English vocabulary could use some improvement, you often seem to fail to find the appropriate words. The difference is particularly noticeable when you switch to Swedish to answer a question, using a much broader range to explain the same thing. You can probably tell when it happens, the important thing would then be to remember what you were trying to say and then try to look up better words/phrases afterwards.

      * You have the bad habit of using "Ähh" and "Öhh" instead of short pauses while talking, it should be easy to correct once you're aware of it.

      * You often seem to suspect that few understood what you said. Rather than asking a question to that effect, you should summarize the main points of what you just said. Doing so is less patronizing for those that did understand, and more useful for those that didn't. Particularly: asking "Do you agree with me?" (which often comes of as "You agree with me." because of lack of intonation) is rarely going to give any useful response. Any student that would have answered "no" to that question would have interrupted you anyway.

      * Mathematical derivations on slides are rather useless. If you want people to understand how a formula is derived (or even how it looks) then you should derive it on the blackboard, where people can follow the train of thought (rather than trying to follow the slide, which is quickly removed). In general, you are easy to understand when using the blackboard, so please take every chance to use it.

      * The slides are often too noisy, there's often too much text per slide for people to be able to extract the important parts while listening to you. You often use complete sentences, that's not necessary for slides. The slides often seem built to be lecture notes (i.e. they often include many details and have a verbosity level suited for undivided attention), rather than to be support for your lecture (i.e. emphasize the main points that you're making). I know that it's easy to construct slides that include everything, because it looks good on the computer. What you need to remember though is that the people at the lecture won't be sitting alone reading the slides on a computer, they will be listening to you while glancing at a (often too small) projection for a few seconds, in an attempt to grasp your main points.





      Hedvig:

      * You might want to sit down and work out the projector issues before class, so that you are confident in how to make it work once in class.

      * Minor detail: The blackboard behind you was full of interesting problems during your lecture, which can be distracting. You didn't have time to erase it this time, but it's worth thinking about.

      * Your slides gave a nice introduction to the problem, but could have used more structure (see above) and less noise (see above). The part that was unclear was that we were jumping between different methods without any clear separator (i.e. it was hard to (during the first part) tell whether the next slide was a continuation of the previous one, or some other method). E.g. you could have tied together the slides on moments while clearly separating them from the rest.

      ---
      Put more emphasis on difficult basic concepts (homogeneous coordinates, Fourier transform and Hough transform for example) and less on vague discussions about applications or simple concepts
      ---
      Since time was limited, the topics were not covered as they were supposed to be. Either the topics can be reduced or the lecture hours per week can be increased


    11. Did you prepare before attending the lectures (by reading relevant sections of the book or notes)?

      1. 0% (0 st) Yes, always.
      2. 24% (4 st) Most often.
      3. 12% (2 st) Sometimes.
      4. 35% (6 st) Rarely.
      5. 24% (4 st) Never.


    12. What do you think of exercise lectures (övningar) ?

      1. 0% (0 st) Very good.
      2. 29% (5 st) Good.
      3. 29% (5 st) Acceptable.
      4. 12% (2 st) Not so good.
      5. 0% (0 st) Poor.
      6. 24% (4 st) Haven't participated.

      Additional comment:

      not many problems were solved/discussed.
      ---
      För få, det borde ha varit fler övningar.
      ---
      Schemakrockar gjorde att jag inte deltog.
      ---
      Kändes lite väl grundläggande när vi löste en endimensionell andragradsekvationen (eller var det tredjegrad? minns inte).
      ---
      It seemed to little with only two, more of them would be appreciated.
      ---
      The exercises did not fit in too well with what was in the course, they were a bit difficult aswell. Would have liked to see some exercises on perspective projections and epipolar geometry aswell.
      ---
      The exercise sessions were insufficient. Furthermore, not all the exercises could be covered in each session. Thus we solved totally 6-8 problems during the entire course during the exercise sessions. This does not really build confidence for facing the exam.


    13. Did you get the support you expected during the lab sessions (are the assistents knowledgable, attendant, friendly)?

      1. 47% (8 st) Sure, yes.
      2. 35% (6 st) Well, yes.
      3. 6% (1 st) Acceptable.
      4. 0% (0 st) Rarely.
      5. 0% (0 st) Never.

      Additional comment:

      i did not have the opportunity to attend many sessions due to clashes with other courses
      ---
      De var mycket hjälpsamma
      ---
      For the last lab, an additional help session would have been useful
      ---
      I didn't actually request any help during lab sessions, but they seemed ok during the reviews.
      ---
      I never asked for help (hence I haven't selected any option).
      ---
      One word: SIMA


    14. How much time did you spend on Lab 1?

      1. 35% (6 st) Less than 5 h.
      2. 41% (7 st) 5-10 h.
      3. 12% (2 st) 10-15 h.
      4. 0% (0 st) 15-20 h.
      5. 6% (1 st) 20-25 h.
      6. 0% (0 st) More than 25 h.
      7. 0% (0 st) Haven't completed Lab 1.

      Additional views on Lab 1:

      not very difficult


    15. How much time did you spend on Lab 2?

      1. 0% (0 st) Less than 5 h.
      2. 41% (7 st) 5-10 h.
      3. 18% (3 st) 10-15 h.
      4. 29% (5 st) 15-20 h.
      5. 6% (1 st) 20-25 h.
      6. 0% (0 st) 25-30 h.
      7. 0% (0 st) More than 30 h.
      8. 0% (0 st) Haven't completed Lab 2.

      Additional views on Lab 2:

      Give more intuitive explanations on the Fourier Transform.


    16. How much time did you spend on Lab 3?

      1. 0% (0 st) Less than 5 h.
      2. 6% (1 st) 5-10 h.
      3. 41% (7 st) 10-15 h.
      4. 24% (4 st) 15-20 h.
      5. 0% (0 st) 20-25 h.
      6. 6% (1 st) 25-30 h.
      7. 0% (0 st) More than 30 h.
      8. 18% (3 st) Haven't completed Lab 3.

      Additional views on Lab 3:

      more crarification / help notes should be given for parts 5 and 6
      ---
      Lite mycket fokus på Houghtransformen trots att det inte var så mycket om det i boken eller på föreläsningarna.
      ---
      please elaborate on the difference between filter2 and conv2.


    17. What do you think of the labs? Were they fruitful? Do they reflect what you believe is important in the course?

      1. 35% (6 st) Very good.
      2. 59% (10 st) Good.
      3. 0% (0 st) Quite good.
      4. 0% (0 st) Not too good.
      5. 0% (0 st) Poor.
      6. 0% (0 st) Haven't finished any lab.

      Additional comments on the labs as a whole:

      The labs could be more stimulating if there was more problem solving questions.
      ---
      They were a bit hard, but very interesting and fun
      ---
      Sometimes you get stuck on some stupid Matlab detail which was unclear in the lab text. This is not good and should not be the focus of the lab.
      ---
      The entire classification concept is still a bit fuzzy. Can it be included in the labs? At least a section in one of the labs.


    18. What do you think about course webpage?

      1. 29% (5 st) Very good.
      2. 47% (8 st) Good.
      3. 12% (2 st) Alright.
      4. 0% (0 st) Not so good.
      5. 0% (0 st) Poor.
      6. 0% (0 st) Never looked at it.

      Additional comment on the web information:

      a link to the exam registration would be useful
      ---
      "news" section could be more useful


    19. How large a fraction of your study time have you dedicated to this course during period 3?

      1. 0% (0 st) Less than 15%.
      2. 12% (2 st) 15-30%.
      3. 47% (8 st) 30-50%.
      4. 29% (5 st) 50-70%.
      5. 6% (1 st) More than 70%.


    20. The course awards you 6 credit points. Is this reasonable compared to other courses?

      1. 0% (0 st) 6p is too much.
      2. 35% (6 st) 6p is reasonable.
      3. 18% (3 st) Should be 7.5p.
      4. 41% (7 st) Should be 9p.


    21. What aspects of the course do you regard as the best, the most fruitful or interesting parts? Why?

      the practical applications at the lab. the hough transform and the image segmentation
      ---
      Själva ämnet. Intressant och känns meningsfullt, man ser att man kan tillämpa sina kunskaper inom datateknik.
      ---
      Line detection is great.

      The topics are generally interesting and the approach is very educational.

      ---
      The lab sessions were really good, since they gave me the opportunity to reinforce the theoretical knowledge by applying it in practice.
      ---
      Laborationerna var överlag bra, speciellt Houghtransformen.
      ---
      Face detection and Hough transform
      ---
      The labs and the exercises, since that's when you get a feel for what you should know.
      ---
      The labs were the best parts because I tend to learn the most when I need to apply theory and not just read about it.
      ---
      Perspective projection and stereo vision
      ---
      Labs. Because they were practical not only theoritical
      ---
      Grey level transformations, Edge detection and linking, high and low pass filtering and general convolution with generic kernels was the most fruitful. This is because I found many direct applications and had a lot of fun playing around with images, not only in MATLAB but also in image editing software like the GIMP which has several cool features including the ability to convolve an image with a generic matrix that you specify. The concepts of classification and morphology were a bit less exciting as direct application was not found/elaborated.


    22. What CAN be done better in the course? How?

      more exercises should be solved
      ---
      One lecture about the biological aspect is too much.

      The practical application of PCA is hard to understand. We don't have real clues about object recognition.

      ---
      Kändes som att delar av kursen inte rymdes i labbarna, tänker framför allt på minkowski-operationerna som kanske kan arbetas in som en del i någon av labbarna?
      ---
      I would appreciate if there were solutions to the old exams and/or more exercises.
      ---
      More labs and help sessions would be great and also try to connect more of the course to labs. I think the examination of this course should rely more on labs than on the test.
      ---
      The intro lecture could have contained more cool examples of what can be done and the examples could have been connected to what we would learn in the course. An example: One could have shown visual odometry and connected that to feature extraction.
      ---
      The course can be done 7.5 ECTS credits or the topics covered can be reduced or lecture hours per week can be increased
      ---
      The course is simply too heavy considering the credit points it gives!! For 4 credit points, the material up to Hough transform is enough, in my opinion.

      Considering what we studied and the time we had to put in 6 or 7.5 credits is what should be given.

      ---
      Give the grading thresholds before the exam. 49 out of 50 points (98%) for an A seems awfully high, especially when a B is only 78%.


    23. Give your general impression of the course as a whole:

      1. 24% (4 st) Very good.
      2. 35% (6 st) Good.
      3. 35% (6 st) Satisfactory.
      4. 0% (0 st) Not so good.
      5. 0% (0 st) Poor.

      Now just press the button below!


    danik@nada.kth.se

    Denna sammanställning har genererats med ACE.