VSS2011 Note

16.526 Blakeslee & McCourt showed that a model that mimic early visual function can actually explain much brightness illusion that long considered as a result of high cognitive function.
We actually did a similar thing in our work on White illusion (Lin & Chen, 2010). Maybe we should try more effects along this line.

26.534 This my favorite presentation on the second day of the meeting Motoyoshi asked why western classical paintings emphasize so much on shading while their Eastern counterparts on outlines. The difference between these two approaches are obvious, with an emphasis on shading, figures in Western paintings tend to have more depth and look more realistic whose those in Eastern painting look flat and artificial. Motoyoshi suggested that the answer may be in the luminance statistics in the respective environment. In the dry Mediterranean region, illumination tends to be highly directional. As a result, there are always a lot shadows and highlights on a figure. On the other hand, in the humid Asian coastal regions, the illumination tend to be diffused. Shadows and highlights are limited. Hence, the artists of these two cultures were both trying their best depicting the reality around them.
If so, I believe that artists from San Francisco would paint like Asians even though their training are western.


22.11 One influential theory about crowding is that the target detection mechanism tend to integrate information from a wide region that covers the flankers. As a result, it is difficult to detect the target without the interference from the flankers. Ester et al. did a target orientation identification experiment in the presence of flanker with different orientation. They found that the error pattern (the distribution of orientation identification error) was not affected by the flanker orientation and thus
concluded that there is no evidence for the spatial summation model for crowding.


23.520. It is known that it is difficult for a computer to identify a face in a captured image if the face is too small or the resolution is too low. Braun & Sinha showed that the human observers can correctly identify a face with a very low resolution. So, the problem, they proposed, is that the computer used only internal features for face identification and ignore the external contour, which contained much information about face identity.

31.22 Ehinger & Oliva asked people to "make photo" in a scene. They found that most people have a large degree of agreement about which view to take. The view point they chose tend to be the place with largest area. Hence, people tend to shot the hall way in a room rather than the walls. people has less consistency in open space , such as parking lots. This study on psychology of photography, I think, would have a large impact.

31.23 It is an interesting idea to have a 3D puffball to fill up a 2D figure before segment it.

34.13 Gheorghiu & Kingdom had their observers adapted either a contour or a contour embedded in a texture and showed that adapting to a texture had less threshold elevation effect than adapting to a contour. In addition, such adaptation effect depends on the relative orientation between the contour and texture elements. This demonstrated a local and a global inhibitory interaction between texture and contour. Their studies may be quite relevant to my own work. I think that I need to pay more attention to their works.

35.12 Geisler proposed a multiple point Baysian statistics that may better summarize the characteristics of a nature image. This approach sound interesting. However, the talk is quite mathematical and lacks many details. Thus, I should pay more attention to the development of this study.

41.13 Sinha et al. studied the re-emergence of several visual cortical functions of a group of patients following their surgery of re-gaining eye sight. They found that regions of the brain selective for facial
responses, including the fusiform facial area (FFA) and occipital facial area
(OFA), developed after the onset of sight. This study is amazing and will have a huge impact on human development and neural plasticity studies.


42.16 Hong & Tong measured the cortical response to form based color filling with SVM. They found all areas from V1 to V4 can tell the difference between two filling-in conditions whereas only V3 and beyond can tell real color from the induced one. Hence, they believe that filling-in is an ability of a higher functions.


43.316 The authors did an experiment that is almost identical to what we reported last year. I talked to both authors. They said that they read our JOV paper (Li & Chen, 2011) only after they submit their abstract.

52.25. It is a super interesting idea that an angry face can be described as nothing but a letter 'X'.

54.12. The authors used the tomographic method, commonly used in CY scan, for pRF computation with fMRI.

54.15. The authors compared two theories for the TMS effect: neural responses suppression and noise increment. They measured psychometric function for orientation discrimination after the observers adapted to either blank screen or a flickering grating and the target threshold was measured with or without applying TMS pulse. The psychometric function shifted to the right after adapting to patterns, indicating a reduce of sensitivity. The TMS pulse shifted the psychometric function for the unadapted condition to the right but the one for the adapted condition to the left. The authors claim this effect rejected noise increment theory. The constant asymptotic level across condition, according to the author, should reject response suppression theory.
I think that the author made several mistakes here. The same asymptotic level for psychometric function has nothing to do with sensitivity. It is about performance, and one cannot have a performance better than perfect. The noise effect should influence the slope of the psychometric function (some raised this comment after the talk), not location. Hence, the location argument is also flawed. The better method to attack the same issue should be noise masking and TvC functions.

53.414 It is a rare visualization study in VSS.

61.11 Petrov compared VEPs for Full field and a sum of quarter fields.I should pay attention to the development of this project.

62.12 Spatiotemporal multi-voxel pattern classification? Worth to try.