• Which the release of FS2020 we see an explosition of activity on the forun and of course we are very happy to see this. But having all questions about FS2020 in one forum becomes a bit messy. So therefore we would like to ask you all to use the following guidelines when posting your questions:

    • Tag FS2020 specific questions with the MSFS2020 tag.
    • Questions about making 3D assets can be posted in the 3D asset design forum. Either post them in the subforum of the modelling tool you use or in the general forum if they are general.
    • Questions about aircraft design can be posted in the Aircraft design forum
    • Questions about airport design can be posted in the FS2020 airport design forum. Once airport development tools have been updated for FS2020 you can post tool speciifc questions in the subforums of those tools as well of course.
    • Questions about terrain design can be posted in the FS2020 terrain design forum.
    • Questions about SimConnect can be posted in the SimConnect forum.

    Any other question that is not specific to an aspect of development or tool can be posted in the General chat forum.

    By following these guidelines we make sure that the forums remain easy to read for everybody and also that the right people can find your post to answer it.

P3D v5 Water detection observations and UE's

Thanks 4 the update. No rush, glad to hear you are to get a little "R&R" (rest and relaxation). We'll be here when you get back. ;)
 
Welcome back!

So I have been reading over your latest release today. Couple of things.

1. sounds like scenProc is now version 3.1 yet the User's Manual says 3.0?

2. Since SVM was in the previous version, I take it this is not the new, improved AI water detection?

If you need someone to test the new AI water detection method, I am ready and willing as I am now back onto developing regions where water will soon come into play. Even if it is still in its raw Python code... I still may be able to test it if you like. Up to you. Which leads me to another question...

3. Do you see any use/advantage of TFE and water detection applied to shorelines, oceans? Have not tried it yet as I have not been developing on the coast but will be eventually.
 
Clutch Cargo said:
1. sounds like scenProc is now version 3.1 yet the User's Manual says 3.0?
Click to expand...
The stable release is now 3.1 and the development release is 3.2. Where do you find the manual that still says 3.0? I have checked and the manuals I have here say 3.1 or 3.2.
Clutch Cargo said:
2. Since SVM was in the previous version, I take it this is not the new, improved AI water detection?
Click to expand...
No, the stable release contains the same features as the development release until now. So it does not have the new water detection yet.
Clutch Cargo said:
If you need someone to test the new AI water detection method, I am ready and willing as I am now back onto developing regions where water will soon come into play. Even if it is still in its raw Python code... I still may be able to test it if you like. Up to you. Which leads me to another question...
Click to expand...
I have just started generating the new training dataset and then in the next days I will train the algorithm again. Hopefully by the end of the week I have a new version so that I can see if it performs better.
Clutch Cargo said:
3. Do you see any use/advantage of TFE and water detection applied to shorelines, oceans? Have not tried it yet as I have not been developing on the coast but will be eventually.
Click to expand...
Do you want to detect where shorelines are or are you just concerned if shoreline water will work as well?
 
Reply to:
1. I re-downloaded the latest and the manual now says version 3.2. Noticed this download was 114MB where in the past it was an "update" of some 22-36MB. I just had it overwrite everything in my scenProc folder.

3. I guess I am asking both, ha! I would like to try and have it detect shorelines as I would think it would be more accurate. In the past sometimes I would just draw straight poly lines to represent the shoreline, where a detection might give more "ebb-&-flows" of the shoreline. I used to use Photoshop to create my shorelines and then used a blendmask to seamlessly move from land into water. The effect looked great however I think since LM changed their water with version 5 (?), the blending looks bad. This will really be my first work back on shorelines for a long time to see whats best. I am willing to try TFE on a shoreline as is to see what the results are first before tweaking to add oceans.

Which leads to another question, I take it TFE would create a water poly (shapefile), out in the ocean as far as the image that is being detected? It will be interesting to see how images next to each other with ocean will look compared to each other as far a blending/smoothness and even blending capabilities. Would you care for a image/samples of an area that has a shoreline? Perhaps I have one that has both the shore and some inlet water/river.
 
Clutch Cargo said:
1. I re-downloaded the latest and the manual now says version 3.2. Noticed this download was 114MB where in the past it was an "update" of some 22-36MB. I just had it overwrite everything in my scenProc folder.
Click to expand...
Correct, now that there is no longer a 32 bit and a 64 bit version, I decided to simplify things and only have one download instead of 3.

I would recommend to install in a clean folder and not over a scenProc 3.1 development release as some files have moved to a different location.
Clutch Cargo said:
3. I guess I am asking both, ha! I would like to try and have it detect shorelines as I would think it would be more accurate. In the past sometimes I would just draw straight poly lines to represent the shoreline, where a detection might give more "ebb-&-flows" of the shoreline. I used to use Photoshop to create my shorelines and then used a blendmask to seamlessly move from land into water. The effect looked great however I think since LM changed their water with version 5 (?), the blending looks bad. This will really be my first work back on shorelines for a long time to see whats best. I am willing to try TFE on a shoreline as is to see what the results are first before tweaking to add oceans.
Click to expand...
I have never tried to detect shorelines. But I guess if you can detect the sea water accurately that would be a good basis to make a blendmask for the shore as well. I guess time will tell how well this works :D.
Clutch Cargo said:
Which leads to another question, I take it TFE would create a water poly (shapefile), out in the ocean as far as the image that is being detected? It will be interesting to see how images next to each other with ocean will look compared to each other as far a blending/smoothness and even blending capabilities. Would you care for a image/samples of an area that has a shoreline? Perhaps I have one that has both the shore and some inlet water/river.
Click to expand...
Once I have the algortihm trained with the new data, I will add an area with a sea shoreline as well in the validation tests that I run. Would be interesting to see how well that goes indeed.
 
Thx, I will perform a clean install of scenProc. Yea, white waves crashing on the shorline or waves hitting rocks should be interesting/challenging.
 
The training run with the new dataset did not produce the improved results as hoped, so I am going to try another approach to see if that helps.
 
"The training run with the new dataset did not produce the improved results as hoped, so I am going to try another approach to see if that helps."

Are we talking about ocean water or just water detection in general? Funny, I am starting a section in more of a desert area (East of the town of Bakersfield, CA, USA), and I did find some water... the water color is brown! This should be interesting to see what TFE makes of this.
 
Not just ocean, any water. It is still missing too much (it's missing more in my Dutch data than in your data actually).

Brown water would be interesting to try, but I guess the NIR channel should help.
 
It's been a month so I thought I would drop a line and see if there has been any progress on the water detection? Seems you were making great strides until you hit a snag with your own local imagery?
 
I have mainly been busy with ModelCoverterX features the last weeks, so not much progress on this.

I did test with some different training data, but the results were not that good. I think the main challenge is to find the right data to train the algorithm robustly. But since training takes a day or two each time (full GPU usage all time, so PC can't be used for other things), that's quite a slow process
 
"I think the main challenge is to find the right data to train the algorithm robustly.

Is there anything I can do to help acquire the type of data you need?
 
If you know of any 4 band imagery that has accurate water data matching the imagery that could be included for the training.
 
I can most certainly search for such data here in the USA. Can you be more specific on the type of data you need? I take it, it is more than simple meta data on an image? Or maybe you can supply a sample of the type of data you need. We can seriously search for this.
 
Basically it's 5 band data. The first four being the R, G, B and NIR data and the last band is water data (255 where there is water and 0 where not). If the water data is not available as raster data, it can also be generated from accurate vector data. Most important is that almost all of the water in the training image is present in the water data used as label.
 
Did some more research. A few more questions. Would a band consisting of the Normalized Differential Water Index (NDWI) work or the fidelity would not be there? Would 3m resolution work or is it better with 50cm resolution?
 
Having NDWI data would not work, it really has to be accurate data of water present or not. And it should have the same resolution as the imagery. Small resolution differences can probably be resized.
 
Would my water polygons or tiffs work? I have several years of hand-created water tiffs that matched up perfectly to 1m images. Since they are black & white could they not be coded to represent water (black, 255) and other (white, 0)?
 
Yes, that would work if they are accurate. Always worth a try :)
 
You must log in or register to reply here.
Back
Top