I have also collected over 50 lenses through the years. I got interested in learning more about micro contrast a while ago but had no idea where to begin as far as establishing a framework to help predict the output. Thank you so much for running these tests! I'm sure it was tedious but it has given me a place to start and a few metrics to track. I'm slowly making my way through your content but have come back to this one a couple times now. Very useful, insightful, and interesting.
@@sneye1 Thanks, I just might take you up on that offer. My eyes don't seem to be quite developed enough to spot the pop consistently at this time, and it would be valuable to have that additional feedback.
The video is hard to watch (to me) since it's so slow and the audio quality is all over the place, but the content is amazing. I wish it showed much more examples
@@sneye1 also please if you find out that the audio is too much on one or the other side and we don't need the directionality, just make it mono not stereo, it is magnitudes better for people using headphones otherwise really interesting videos, thank you, your english is really good and clear to understand, don't be too harsh on yourself about it
Hi, I commented on one of your other videos. Thanks for pointing me to this one. At 13:03 you mention the 3D score prediction model, what factors go into this? You mentioned that it was fitted by multiple regression, and I believe you said you included one product term, but what were the weights of the model? In addition, where to the p values come from? Is this from cross validation? If you have seven terms in your model and 50 data points it is quite easy to fit a model that predicts even a random distribution. And the testing that you show later is using the model to predict its own data if I'm not mistaken. e.g., you use the partition tree, which is fitted from the data you acquired from the images of the plant and books to regurgitate the scores you gave to the images. In addition it doesn't seem that focal length was considered as a variable in your model, is this because you don't believe it is important? Shorter focal length lenses need more correction since the light is bent more and deviate from paraxial lenses more than longer lenses. Because of this shorter focal length lenses need more corrective elements to achieve good image quality so are likely to contain more aspherics and ED glass. Further, shorter focal length lenses, for the same image size, produce less blur in the background - which also seems to be important in 3D pop. Thanks for putting together this video series. I'll watch them all through carefully. I am also interested in 3D pop and I was planning on putting together a similar test to this. I was planning on asking a range of people to score images for 3D pop level, in order to take my opinion out of the scoring.
Please do that! Will be interesting. Focal length was taken as a variable but did not correlate with the 3D scores. Importantly, I ran a stepwise process to eliminate interacting variables from the regression. The model is not yet validated with a new set of observations. Thanks for the scientific critique.
@@sneye1 It's very interesting. We certainly need more data. I do like that you made quantifiable measurements, so many blogs and videos make vague statements it is nice to see an actual graph.
@@sneye1 Thank you! Hope to find some in the 22-28mm range Ahh, I see you reccommend the Nikon 20-35mm. I still don't quite understand what the amount of elements really does in terms of poppiness - I initially thought less elements mean more 3d pop, but the 20-35mm has rather many elements (14) and still pops.
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
I have also collected over 50 lenses through the years. I got interested in learning more about micro contrast a while ago but had no idea where to begin as far as establishing a framework to help predict the output. Thank you so much for running these tests! I'm sure it was tedious but it has given me a place to start and a few metrics to track.
I'm slowly making my way through your content but have come back to this one a couple times now. Very useful, insightful, and interesting.
You are most welcome. Will be interesting to follow your results. If you wish, I can participate in scoring your test photos.
@@sneye1 Thanks, I just might take you up on that offer. My eyes don't seem to be quite developed enough to spot the pop consistently at this time, and it would be valuable to have that additional feedback.
The video is hard to watch (to me) since it's so slow and the audio quality is all over the place, but the content is amazing. I wish it showed much more examples
Thank you. I'm a slow talker, I know. Talking to the wall does not help either. Will try to improve...
@@sneye1 also please if you find out that the audio is too much on one or the other side and we don't need the directionality, just make it mono not stereo, it is magnitudes better for people using headphones
otherwise really interesting videos, thank you, your english is really good and clear to understand, don't be too harsh on yourself about it
@@sneye1 it's fine. The content is still great. I'd love to see more about 3D pop.
Hi, I commented on one of your other videos. Thanks for pointing me to this one. At 13:03 you mention the 3D score prediction model, what factors go into this? You mentioned that it was fitted by multiple regression, and I believe you said you included one product term, but what were the weights of the model?
In addition, where to the p values come from? Is this from cross validation?
If you have seven terms in your model and 50 data points it is quite easy to fit a model that predicts even a random distribution. And the testing that you show later is using the model to predict its own data if I'm not mistaken. e.g., you use the partition tree, which is fitted from the data you acquired from the images of the plant and books to regurgitate the scores you gave to the images.
In addition it doesn't seem that focal length was considered as a variable in your model, is this because you don't believe it is important? Shorter focal length lenses need more correction since the light is bent more and deviate from paraxial lenses more than longer lenses. Because of this shorter focal length lenses need more corrective elements to achieve good image quality so are likely to contain more aspherics and ED glass. Further, shorter focal length lenses, for the same image size, produce less blur in the background - which also seems to be important in 3D pop.
Thanks for putting together this video series. I'll watch them all through carefully. I am also interested in 3D pop and I was planning on putting together a similar test to this. I was planning on asking a range of people to score images for 3D pop level, in order to take my opinion out of the scoring.
Please do that! Will be interesting.
Focal length was taken as a variable but did not correlate with the 3D scores.
Importantly, I ran a stepwise process to eliminate interacting variables from the regression.
The model is not yet validated with a new set of observations.
Thanks for the scientific critique.
@@sneye1 It's very interesting. We certainly need more data. I do like that you made quantifiable measurements, so many blogs and videos make vague statements it is nice to see an actual graph.
Wow! Do you, perhaps, have a large list of all lenses that are more on the poppy side?
Look through the older videos on this channel. Especially 4 to 7.
@@sneye1 Thank you! Hope to find some in the 22-28mm range
Ahh, I see you reccommend the Nikon 20-35mm.
I still don't quite understand what the amount of elements really does in terms of poppiness -
I initially thought less elements mean more 3d pop, but the 20-35mm has rather many elements (14) and still pops.