The second paragraph on the page describes line pairs per millimeter (lp/mm) in the following way, which gives a misleading impression of the relationship between lp/mm and magnification (distance and focal length):
Resolution is often expressed in terms of the sensor's photosite counts or the captured image's pixel counts, but at best these measurements only tell us about one limiting factor. A better, empirical measurement, which reflects all characteristics of the lens, sensor, and setup, is called line pairs per millimeter (lp/mm). This means the maximum density of black-on-white lines that the lens and camera can resolve, in a given setup. At any higher density than this, the lines in the captured image blur together. Note that lp/ mm varies with the subject's distance and the lens's settings, including the focal length (optical zoom) of a zoom lens. When you approach the subject, zoom in, or swap out a short lens for a long lens, the system should of course capture more detail! However, lp/mm does not vary when you crop (digitally zoom) a captured image.
Actually, lp/mm refers to millimeters on the image sensor's surface (not on the subject's surface). For example, a microfilm system with a very sharp lens might resolve 250 lp/mm under optimal conditions. If the lens is focused at 1:5 magnification (one-fifth of real-life size), the microfilm system can reproduce details that are just 20 μm big when measured on the subject's surface:
1 mm / 250 / (1/5) = 20 μm
Alternatively, if the lens is focused at 1:30 (one-thirtieth of real-life sized) magnification, the system can reproduce details that are 120 μm big when measured on the subject's surface:
1 mm / 250 / (1/30) = 120 μm
Ideally, lp/mm would not vary with focal length or distance to the subject. However, a real lens is optimized for a certain range of magnifications or distances to the subject. Typically, long lenses and macro lenses (which are optimized for magnification ratios close to 1:1 or real-life size) do tend to have higher resolution in terms of lp/mm. However, the relationship is not as trivial as the description on page 9 suggests.
Erratum reported by Joseph Howse on July 19th, 2016.
As reported by a buyer of the book, page 189 contains several errata which should be changed, in order to make the explanation more cohesive.
-
Before the line
Using the software is quite straightforward:, there is the mentioning of the opencv integrated version of the tool. However if OpenCV is installed correctly it is a system tool and thus there needs to be no slash before the executable name. The line/opencv_annotation -images <folder location> -annotations <output file>should thus be replaced byopencv_annotation -images <folder location> -annotations <output file>. -
Under number 1 at the bottom of the page there is the
cmakemakecommand which is wrong. It should be replaced by
cmake CMakeLists.txt
make
- Finally the at number 1 there is already the annotation command, which needs files that are being generated in later fases. Therefore the line could be removed in order to keep it more straightforward, since the command is returning at bullet 5.
Erratum reported by John Allen on March 19th, 2016.
At the top of the page, the following explanation can be found for this parameter:
This is the threshold that defines how much of your negative samples need to be classified as negatives before the boosting process should stop adding weak classifiers to the current stage. The default value is 0.5 and ensures that a stage of weak classifier will only do slightly better than random guessing on the negative samples. Increasing this value too much could lead to a single stage that already filters out most of your given windows, resulting in a very slow model at detection time due to the vast amount of features that need to be validated for each window. This will simply remove the large advantage of the concept of early window rejection.
In this explanation 2 things should be changed:
- The sentence
This is the threshold that defines how much of your negative samples need to be classified as negatives before the boosting process should stop adding weak classifiers to the current stage.should be changed toThis is the threshold that defines how much of your negative samples are allowed to be classified as positives before the boosting process should stop adding weak classifiers to the current stage.. - Then correspondingly the word
Increasingshould be replaced byDecreasing.
Erratum reported by LorenaGdL on April 10th, 2016.
Near the end of the page, the following line can be found
This can take some time according to the size of your model and according to the amount of training samples, especially when knowing that a model of 24x24 pixels can yield more than 16,000 features.
In this line, the total number of features should be 160,000 instead of 16,000.
Erratum reported by LorenaGdL on April 10th, 2016.
In the image contained inside page 208, visualising the calculation of a HAAR wavelet like feature, there is a mistake in the formula.
The current formula is stating
feature = area(A) + area(C) - area(C)
while it should be
feature = area(A) + area(C) - area(B)
to correctly represent the image above.
Erratum reported by Can Ergun on November 22nd, 2015.
There were several issues in the precision recall calculation code which have been fixed through the following PR. Things that were fixed with this update:
- Fixed the fact of double TP calculation during the second loop
- Fixed FN calculation
- Removed the 50% larger check, which is implicitly done with the 50% overlap check
- Applied some early loop skipping when not needed = faster processing
- Added some debug output for future use and testing purposes
The new code can be found here.
Errata reported by Nuno Pessanha Santos on July 19th, 2016.