Seems like you're doing everything right to me, however, the PointCloudSplitter might be closer to what you actually want. It should work in a similar way to the Filter here but in this case it will create an attribute so you can see what the actual value is.
Have you been able to inspect the pointcloud where classification_flags != 0 to see the values?
If you're able to show that LASTools is providing different results to FME based on this number then this seems like a potential bug.
Seems like you're doing everything right to me, however, the PointCloudSplitter might be closer to what you actually want. It should work in a similar way to the Filter here but in this case it will create an attribute so you can see what the actual value is.
Have you been able to inspect the pointcloud where classification_flags != 0 to see the values?
If you're able to show that LASTools is providing different results to FME based on this number then this seems like a potential bug.
Thank you that was a good idea to try with classification_flags on the Splitter. Interesting results.
So basically the filter has only two output port I think, flag set to 0 or not set to 0. It only checks if a flag has been set on a point.
Filter:
@Component(classification_flags)==0: 1,937,761 points
@Component(classification_flags)!=0: 11,620,899 points
PointCloudPropertyExtractor (max number of flags) shows 8 though, which is weird because it does not output anything at @Component(classification_flags)==8
Potential bug there.
Splitter:
Split on classification_flags, left empty to keep all. Three outputs, split value 0, 4 and 8
classification_flag 0: 1,937,761 points
classification_flag 4: 1,662,158 points
classification_flag 8: 9,958,741 points
filter==0 is the same as splitter 0 (no flag set on a point)
filter!=0 is the same as spliter (4) + splitter (8)
In a LAS viewer, 0 is flag = none, (4) is Withheld, (8) is overlap.
I tested on another dataset, which comes with additional flag values. Splitter (2) is key point, splitter (10) outputs the points flagged with both 'Overlap,Key-point'. (0) is still none, (8) is still Overlap.
Are those value documented somewhere? I thought it would only be 0,1,2,3 but it seems like there are FME specific values.
Thank you that was a good idea to try with classification_flags on the Splitter. Interesting results.
So basically the filter has only two output port I think, flag set to 0 or not set to 0. It only checks if a flag has been set on a point.
Filter:
@Component(classification_flags)==0: 1,937,761 points
@Component(classification_flags)!=0: 11,620,899 points
PointCloudPropertyExtractor (max number of flags) shows 8 though, which is weird because it does not output anything at @Component(classification_flags)==8
Potential bug there.
Splitter:
Split on classification_flags, left empty to keep all. Three outputs, split value 0, 4 and 8
classification_flag 0: 1,937,761 points
classification_flag 4: 1,662,158 points
classification_flag 8: 9,958,741 points
filter==0 is the same as splitter 0 (no flag set on a point)
filter!=0 is the same as spliter (4) + splitter (8)
In a LAS viewer, 0 is flag = none, (4) is Withheld, (8) is overlap.
I tested on another dataset, which comes with additional flag values. Splitter (2) is key point, splitter (10) outputs the points flagged with both 'Overlap,Key-point'. (0) is still none, (8) is still Overlap.
Are those value documented somewhere? I thought it would only be 0,1,2,3 but it seems like there are FME specific values.
OK so yep, seem a little bit like a bug to me, however, I can see what is happening. Here's a tables from the specs of LAS 1.4.
Note that in fact the number actually represents the bit position- Not exactly an actual number. FME reads/interprets the 4 bits of binary (0000) as a Uint8.
Here's a link showing the conversion between normal numbers (decimal) and Uint8: https://makeapppie.com/2018/04/25/tips-why-use-int8/ (take a look at the table)
In particular you can see that the position of the 1 in the binary block is what gives the classification_flags it's value.
So, 0001 is a 1 in FME, 0010 is a 2 in FME, 0100 is a 4 in FME, 1000 is an 8 in FME. These 4 sets represent the four different types. You can see the 1 moves position from index 0 to index 3.
A combination, like '1010', represents two types (8 and 2 in FME) but is actually interpreted by FME as a 10.
From the link you can see that there are 16 different theoretical combinations depending on which bits are set to 1. FME will return a number from 0 to 15. This can be easily mapped with an AttributeValueMapper to a proper string.
Thank you that was a good idea to try with classification_flags on the Splitter. Interesting results.
So basically the filter has only two output port I think, flag set to 0 or not set to 0. It only checks if a flag has been set on a point.
Filter:
@Component(classification_flags)==0: 1,937,761 points
@Component(classification_flags)!=0: 11,620,899 points
PointCloudPropertyExtractor (max number of flags) shows 8 though, which is weird because it does not output anything at @Component(classification_flags)==8
Potential bug there.
Splitter:
Split on classification_flags, left empty to keep all. Three outputs, split value 0, 4 and 8
classification_flag 0: 1,937,761 points
classification_flag 4: 1,662,158 points
classification_flag 8: 9,958,741 points
filter==0 is the same as splitter 0 (no flag set on a point)
filter!=0 is the same as spliter (4) + splitter (8)
In a LAS viewer, 0 is flag = none, (4) is Withheld, (8) is overlap.
I tested on another dataset, which comes with additional flag values. Splitter (2) is key point, splitter (10) outputs the points flagged with both 'Overlap,Key-point'. (0) is still none, (8) is still Overlap.
Are those value documented somewhere? I thought it would only be 0,1,2,3 but it seems like there are FME specific values.
but we should let @nampreetatsafe know about this so that Safe can update the documentation on the LAS reader about the classificaiton_flags
Thank you that was a good idea to try with classification_flags on the Splitter. Interesting results.
So basically the filter has only two output port I think, flag set to 0 or not set to 0. It only checks if a flag has been set on a point.
Filter:
@Component(classification_flags)==0: 1,937,761 points
@Component(classification_flags)!=0: 11,620,899 points
PointCloudPropertyExtractor (max number of flags) shows 8 though, which is weird because it does not output anything at @Component(classification_flags)==8
Potential bug there.
Splitter:
Split on classification_flags, left empty to keep all. Three outputs, split value 0, 4 and 8
classification_flag 0: 1,937,761 points
classification_flag 4: 1,662,158 points
classification_flag 8: 9,958,741 points
filter==0 is the same as splitter 0 (no flag set on a point)
filter!=0 is the same as spliter (4) + splitter (8)
In a LAS viewer, 0 is flag = none, (4) is Withheld, (8) is overlap.
I tested on another dataset, which comes with additional flag values. Splitter (2) is key point, splitter (10) outputs the points flagged with both 'Overlap,Key-point'. (0) is still none, (8) is still Overlap.
Are those value documented somewhere? I thought it would only be 0,1,2,3 but it seems like there are FME specific values.
oh woah, thanks for this, it makes complete sense!
The PropertyExtractor did show Uint8 and yeah, the different combination binary/decimal are matching. Thank you
Thank you that was a good idea to try with classification_flags on the Splitter. Interesting results.
So basically the filter has only two output port I think, flag set to 0 or not set to 0. It only checks if a flag has been set on a point.
Filter:
@Component(classification_flags)==0: 1,937,761 points
@Component(classification_flags)!=0: 11,620,899 points
PointCloudPropertyExtractor (max number of flags) shows 8 though, which is weird because it does not output anything at @Component(classification_flags)==8
Potential bug there.
Splitter:
Split on classification_flags, left empty to keep all. Three outputs, split value 0, 4 and 8
classification_flag 0: 1,937,761 points
classification_flag 4: 1,662,158 points
classification_flag 8: 9,958,741 points
filter==0 is the same as splitter 0 (no flag set on a point)
filter!=0 is the same as spliter (4) + splitter (8)
In a LAS viewer, 0 is flag = none, (4) is Withheld, (8) is overlap.
I tested on another dataset, which comes with additional flag values. Splitter (2) is key point, splitter (10) outputs the points flagged with both 'Overlap,Key-point'. (0) is still none, (8) is still Overlap.
Are those value documented somewhere? I thought it would only be 0,1,2,3 but it seems like there are FME specific values.
For whoever read this in the future, here is the result after applying flags 'manually' with lastools to a sample, here is how FME and the Splitter return it:
synthetic: classification_flags == 1
key point: classification_flags == 2
withheld: classification_flags==4
overlap: classification_flags==8
Is there a trick on how to read and count points with a classification flag?