curvy slabs

[Anonymous]

Hello,

assume you have a 40 years old archive of about 500 hand made projects. As they are both valuable and profitable, you want to convert them into a digital archive. About 4/5 of the drawings consist of projections that can be calculated on the fly by a BIM CAD, so the problem reduces to importing the 2d plots. You take someone in your staff and ask him to do this accurately. He takes an old project, make a tiff image out of it, then read the image with the CAD and... now what? The person comes back to you with a pesky problem: the slabs are curvy, hand drawn, and the conversion turns out to be really difficult and less than accurate. You are not allowed to change the original shapes: the digital version must be an accurate 1:1 import of the original. I recollect the similar problem occurring in the 80', when the CAD could only draw straight lines and we had to give up with the very idea of using a CAD. Twenty+ years later, the situation has improved, so much that you can draw a curvy slab directly on the CAD, but the problem is still pending when importing old projects. There is indeed a difference between drawing directly in CAD and importing from an old project, and the 1:1 conversion is a must. The conversion of hundreds of project makes the case for a rapid method. Then my question is, how would you solve this problem with AC10?

[Dwight]

Vector scan programs only define contrast edges, not drawing intent. Printing induces ambiguities that these applications cannot overcome to the detail architects need.

[Anonymous]

Mhhh... looks like we should give up with the very idea of automatic 1:1 vectorization for CAD.

I looked at SketchUP again, which is so easy with straight lines and regular half circles. But when it comes to free curves, I have trouble in matching vectors with the underlying raster. My own eyes are in pain (-6 of myopia).

[Dwight]

Some things that should be automatic, aren't.

It always depends on your standards. How close do you need to be?

I figure you'd need to scan a 6x0 rapidograph on mylar film at 300 dpi to get an acceptable auto-trace, but even then, inadvertent stretching of the film and imperfect placement will mean the scanlines will occasionally misalign one or more pixels for a crooked interpolation.

Man-yule is best.

[Thomas Holm]

Seems like my masters are chiming in to save my skin....

Bob. The truth is, you will never get an accurate model from the process you describe. If you're ready to accept errors, then OK. But small errors tend to add up, or worse, multiply.

In addition to what my colleagues say, I'd like to add the obvious fact that a line drawing is a symbol. When we draft manually, we utilize this fact. Things are supposed to look good and clear, and roughly to dimensions given. But we also state (usually in the drawing's description) that when dimensions are given as numbers, they count. Dimensions taken by measuring the drawing with a ruler are not valid. There's a reason for that!

Also, I've used Illustrator's vectorizing, I just say good luck! It will even save as dwg which you can import into Archicad lines, not just as an uneditable pdf. It is very good. For logos.

But not good enough for this purpose. If you want vectorizing as good as it gets, you will need a specialist - a service bureau. Take Matthew's and Dwights advice! Also, get some professional Archicad training! Money spent on that will pay for itself, many times over what you spend on useless vectorizing!

[Dwight]

The parallel is renovating an old house. At some point you make new over what is a crooked original. Twenty coats of paint and the dimensions aren't real anymore.

[Anonymous]

>Also, get some professional Archicad training!

This is past the deadline, as we are near the decision of Archicad vs Revit, stepping away from AutoCAD. But the specific problem of automatic vectorization from a printed project has nothing to do with an individual's expertise in using the tool named ArchiCAD. I disagree with the very claim that automatic vectorization is necessarily bad. It just happens that the present vectorizers are less than perfect, just like OCRs are, but some are better than others, and technical research is making progress, just like it did with stone-age AutoCAD vs ArchiCAD's BIM. I am free from the dogma that things must stay as they are at this time. I understand that the main argument against proper vectorization is, that it would take work away from human professionals. I just belong to a different school.

Bob

[Thomas Holm]

jdk wrote:
I understand that the main argument against proper vectorization is, that it would take work away from human professionals.

Bob,

I'm sorry, but if THAT is what you've read in our posts, then you've understood nothing!

Even if I'm old, I've used CAD since it was new, and have had - and still have - high hopes for automation. I've used vectorization too. It's fine for what it does, but it simply doesn't give you a good base for BIM models, when what it does is interprete manual, copied and scanned drawings And the reasons for this are inherent in the process.

It's not the vectorization per se. We're telling you that today it is faster to use a trained person to input the data, than to use that same person to correct erroneous input from a program.

My conclusion might be different if you were speaking about using data from on-site point cloud laser measurements. But you are not.

If you don't want to listen, you've got a problem. Not our problem though. Take off those pink glasses!

[Anonymous]

jdk wrote:
I disagree with the very claim that automatic vectorization is necessarily bad. It just happens that the present vectorizers are less than perfect, just like OCRs are, but some are better than others, and technical research is making progress

Bob,

You are talking to some of the most advanced, technology friendly people anywhere. We all got into this to find ways to automate the drudge work so that we can spend more time improving the quality of the work (design). We are also realistic about what is possible.

My physics professor once pointed out that if you have paced off the diameter of a circle you can't get a more accurate value for the circumference by adding decimal places to your value for pi.

As far as vectorization goes it is fundamentally limited by the quality of the information going in (the GIGO principle in computer science). Line work on paper (or ink on mylar) simply isn't accurate enough to produce a proper base for a building model. Now it may be that someday computers will actually be able to read drawings, but I wouldn't hold my breath waiting for this development. Especially considering that most of our highly educated clients don't understand architectural drawings either.

I can imagine two developments that might improve the situation.

1. BIM software could become more accommodating of errors and discrepancies. This was sort of true with integer based CAD since the field of available points was more limited. In the floating point world it would require some built in way to keep things tidy within some defined margin of error and prevent small errors accumulating or showing up as gross anomalies. This could be an advance if well implemented, but any time the software makes assumptions and automatic adjustments I get worried about how well the programmers have anticipated my needs.

2. Vectorization with interpretation. This would require some advanced heuristics to rectify nearly right angles, normalize small fractions, and so on. This would be similar to adding a spell checker to OCR to improve accuracy, only much more complex.

The comparison to OCR is probably apt if you also consider that vectorization is a much more complex task. Considering that OCR has a vastly larger market and is still not entirely worked out I think we can expect to wait a very long time before we see any acceptable results in this area for our little corner of the marketplace.

There are some things that computers just aren't very good at and recognizing complex patterns of information represented graphically is one of them. There are some ways that John Henry still beats the steam drill.

[Anonymous]

I enjoyed the discussion, and I thank you for it. We still have the problem of how to vectorize cury slabs, by hand or otherwise. The type of curves we have are difficult to replicate, firstly because they were hand drawn without any geometrical reference, such as foucs points for ellipses. The 1:1 replication of these curves is a challenge. Perhaps the most rapid solution is to use a very-high resolution tablet, with a digital pen, and go over the lines by hand.

>I've used vectorization too. It's fine for what it does, but it simply doesn't give you a good base for BIM models, when what it does is interprete manual, copied and scanned drawings And the reasons for this are inherent in the process.

As I go along with most of what I've red in the messages, I conclude that we misunderstood each other. I was not wishing for a magic tool that builds and ArchiCAD model out of a printed project; it would be wonderful to have, but is clearly not around at this time. It is clear that one has to rebuild the model by hand, and I cannot recollect anything I've said that might have triggered the doubt. What I have said is, that a magic wand (similar to the one that already exists in ArchiCAD) could help in the process of vectorizing curvy slabs. Another idea is to use interpolating points with concavities, instead of re-tracing the curves by hand, which process is error prone. I've seen many CAD designs, but none of them had this problem. CAD is good for straight lines and regular curves, but the sort of curves we have are a challenge.

[Anonymous]

jdk wrote:
I enjoyed the discussion, and I thank you for it. We still have the problem of how to vectorize cury slabs, by hand or otherwise. The type of curves we have are difficult to replicate, firstly because they were hand drawn without any geometrical reference, such as foucs points for ellipses. The 1:1 replication of these curves is a challenge. Perhaps the most rapid solution is to use a very-high resolution tablet, with a digital pen, and go over the lines by hand.

For tracing freehand curves I use the spline tool. It is very easy to use with a minimum of control points. If your curves do not have a strict defining geometry then close enough probably is. You can then magic wand to the spline to approximate the curve with arcs for the model elements (they don't like splines).