DroneDeploy 3D Annotations Hijacked to Processing Software Comparisons

[adm_geomatics]

***Obviously this thread went its own way so feel free to comment on either topic, lol.

Been chatting in the Terrain Awareness thread which brought to mind another feature that is fairly recent that I am not sure how many people are using. Have you used the annotations feature on the 3D model? If so, what do you think? Do you have any issues? I usually digitize my areas in 2D with the Elevation layer on so that I can see well-defined breaklines and then go into 3D to make sure that I don't have any red/blue that is not needed for the situation. The 3D navigates nicely, but I do have some issue manipulating the 3D vertices as they seem hard to pick.



A tip is to change the Elevation slider to isolate the elevation range of the area you are in. This site has a pretty good elevation change so the colors are spread further and there isn't as much definition in each area.



With annotations... This is where the 3D is valuable as I need to eliminate the blue to get as close to true export as possible. Within time consumption reason...

 

[clolsonus]

I was bummed out that I couldn't share my drone deploy maps with fellow team members and let them have access to add annotations. I was told I would need to buy an entire seat for each team member that wanted access just to do annotations. I was the only one flying and making the maps and consuming big resources ... I just wanted to share the maps with a subject matter expert to go through them and annotate them. (I could share them, they just couldn't annotate them.) I was told it would be a violation to share my password to work around this licensing limitation .... it would have been about $2k per person at our educational institute rates.

We discovered these professional tools are great when they do exactly what you need, but frustrating when our use case strayed a little outside the norm (which is almost always because we are a research lab.) We tried exporting the maps as a geotiff so we could share them that way, but had to cut resolution tremendously or our maps would 'exceed the limitations'. Our use case was finding a needle in the haystack so we needed every bit of resolution the original images offered.

I did feel (subjectively) that drone deploy produced a higher quality result than pix4d for our types of areas (agricultural fields, and forests--sometimes with significant terrain.) Pix4d also can go pretty bonkers if it goofs up an area, drone deploy seemed to be smarter about knowing what it could or couldn't stitch and avoided big goof ups. Some of that is on me for not always having great data, but due to operational constrains, limited access points, elevation changes, etc., I sometimes get forced into not being able to fly the ideal mission profile or getting inconsistent overlaps (too much in the low lying areas, not enough at the high points.) Anyways, lots of fun, still learning as I go .....

 

[adm_geomatics]

I was bummed out that I couldn't share my drone deploy maps with fellow team members and let them have access to add annotations. I was told I would need to buy an entire seat for each team member that wanted access just to do annotations. I was the only one flying and making the maps and consuming big resources ... I just wanted to share the maps with a subject matter expert to go through them and annotate them. (I could share them, they just couldn't annotate them.) I was told it would be a violation to share my password to work around this licensing limitation .... it would have been about $2k per person at our educational institute rates.

We discovered these professional tools are great when they do exactly what you need, but frustrating when our use case strayed a little outside the norm (which is almost always because we are a research lab.) We tried exporting the maps as a geotiff so we could share them that way, but had to cut resolution tremendously or our maps would 'exceed the limitations'. Our use case was finding a needle in the haystack so we needed every bit of resolution the original images offered.

I did feel (subjectively) that drone deploy produced a higher quality result than pix4d for our types of areas (agricultural fields, and forests--sometimes with significant terrain.) Pix4d also can go pretty bonkers if it goofs up an area, drone deploy seemed to be smarter about knowing what it could or couldn't stitch and avoided big goof ups. Some of that is on me for not always having great data, but due to operational constrains, limited access points, elevation changes, etc., I sometimes get forced into not being able to fly the ideal mission profile or getting inconsistent overlaps (too much in the low lying areas, not enough at the high points.) Anyways, lots of fun, still learning as I go .....

What level of plan are you all running on? Have you thought about collaborating using QGIS? I think that would be a really good move since you are in research. You can definitely work with whatever high level you can get out of DroneDeploy. It will also get you a little deeper into the GIS side of the world which for annotating is much more powerful than DD. The only thing that you really can do there easily is volumetrics, but there are many other program of the CAD nature that do that better than DD as well. PM me if I can help.

 

[clolsonus]

What level of plan are you all running on? Have you thought about collaborating using QGIS? I think that would be a really good move since you are in research. You can definitely work with whatever high level you can get out of DroneDeploy. It will also get you a little deeper into the GIS side of the world which for annotating is much more powerful than DD. The only thing that you really can do there easily is volumetrics, but there are many other program of the CAD nature that do that better than DD as well. PM me if I can help.

My understanding is that our educational license was ball park similar to their enterprise option (it's been a year or two now so my memory is a bit foggy.) We paid for one year, but when that year expired we let our subscription lapse.

The big problem is that the geotiff export option from drone deploy wouldn't let us output the map at full resolution. In fact we had to cut the resolution down by a factor of 8 before I could get the map to successfully export without bombing with a "sorry, too big" message.

Our use-case is flying low level, capturing the highest resolution imagery possible, and then looking for specific colored berries (oriental bittersweet) in forested areas. So cutting our resolution down by a factor of 8 when the original resolution is borderline for what we were trying to do was a no-go.

Not to steer your thread away from it's original intent, but in the end we developed our own tools that would 'perfectly' fit all the original images together into a mosaic map and then let the operator navigate around (pan/zoom) and also then for any specific point, be able to select which (of up to 9 images that cover the spot) to show as the primary image. This ends up working so much better for us. Our target vine can grow up tree trunks and be totally hidden by the canopy in an ortho-map ... but when we look at the individual images in their map orientation/scale/context we can clearly see the target vines. Also, our tools preserve all the max original detail of the original photos because it's drawing out the original photos to produce the map.

There are tons of different use cases out there, and pix4d and drone deploy (and others) are great software ... I'm not intending my comments to sound negative at all ... I'm just talking about our one frustration with drone deploy annotations for our "sharing" use case (which was one of the reasons among many we developed our own software tools.) The tools are all open-source by the way, I'm [slowly] working my way towards packaging them in a way that a non-computer expert could run them. But if you are a person who isn't afraid of installing python packages (some from git) and possibly recompiling opencv from source code to turn on SIFT support, this could be just the thing for you.

 

[adm_geomatics]

My understanding is that our educational license was ball park similar to their enterprise option (it's been a year or two now so my memory is a bit foggy.) We paid for one year, but when that year expired we let our subscription lapse.

The big problem is that the geotiff export option from drone deploy wouldn't let us output the map at full resolution. In fact we had to cut the resolution down by a factor of 8 before I could get the map to successfully export without bombing with a "sorry, too big" message.

Our use-case is flying low level, capturing the highest resolution imagery possible, and then looking for specific colored berries (oriental bittersweet) in forested areas. So cutting our resolution down by a factor of 8 when the original resolution is borderline for what we were trying to do was a no-go.

Not to steer your thread away from it's original intent, but in the end we developed our own tools that would 'perfectly' fit all the original images together into a mosaic map and then let the operator navigate around (pan/zoom) and also then for any specific point, be able to select which (of up to 9 images that cover the spot) to show as the primary image. This ends up working so much better for us. Our target vine can grow up tree trunks and be totally hidden by the canopy in an ortho-map ... but when we look at the individual images in their map orientation/scale/context we can clearly see the target vines. Also, our tools preserve all the max original detail of the original photos because it's drawing out the original photos to produce the map.

There are tons of different use cases out there, and pix4d and drone deploy (and others) are great software ... I'm not intending my comments to sound negative at all ... I'm just talking about our one frustration with drone deploy annotations for our "sharing" use case (which was one of the reasons among many we developed our own software tools.) The tools are all open-source by the way, I'm [slowly] working my way towards packaging them in a way that a non-computer expert could run them. But if you are a person who isn't afraid of installing python packages (some from git) and possibly recompiling opencv from source code to turn on SIFT support, this could be just the thing for you.

That's all very interesting! I think there might have been some confusion on the licensing. When was this? We've been an Enterprise customer for 3 years and I would love to try some of your data if that is acceptable. I'd also like to chat with DroneDeploy about what kind of package they were giving you guys because depending upon what altitude you were flying we should be able to get you down to 0.5 inch pretty easily. Either way I still think QGIS would be a pretty amazing platform for you guys as it also accept Python scripts so much of what you have already written may be applicable. If you're interested in chatting or sharing some sample data then PM me and I would be happy to spend some hobby time on it. All that said your sharing options should have been much more open if you were on anything close to an Enterprise plan.

 

[clolsonus]

I don't recall exactly, but I think we purchased our drone deploy (educational) license about 2.5 years ago. From memory I understood we were getting most of the $3k enterprise functionality (1 seat) for about $2k. The group we are working with supposedly does a lot of arcgis, but I've never had access to a license that I know of. I need to ask around our university, but everything is always so silo'd you need to be in the right place at the right time to know about much of the other work going on here.

For us it wasn't so much about inches per pixel, but the process of generating a dense point cloud and then ortho projecting that onto a flat map loses a lot of the original information in the original image set. Much of that lost information is useful for looking at an area of interest from the slightly different angles.

Anyway, got to run here, but happy to talk more about this ...

Thanks,

Curt.

 

[adm_geomatics]

I don't recall exactly, but I think we purchased our drone deploy (educational) license about 2.5 years ago. From memory I understood we were getting most of the $3k enterprise functionality (1 seat) for about $2k. The group we are working with supposedly does a lot of arcgis, but I've never had access to a license that I know of. I need to ask around our university, but everything is always so silo'd you need to be in the right place at the right time to know about much of the other work going on here.

For us it wasn't so much about inches per pixel, but the process of generating a dense point cloud and then ortho projecting that onto a flat map loses a lot of the original information in the original image set. Much of that lost information is useful for looking at an area of interest from the slightly different angles.

Anyway, got to run here, but happy to talk more about this ...

Thanks,

Curt.

Yeah, allot has changed in the last 2 years. The platform is probably allot more applicable now. I don't work for them, but am one of their customer advocates and operate the forum so I would be happy to help, especially a cause such as this.

 

[clolsonus]

We are at the tail end of our project funding, so restarting a $2k/year subscription probably isn't an option for our particular situation. It's a bummer that all our previous work and maps stored in the drone deploy sphere automatically went to low-res demo mode when our subscription lapsed ... that is the non-permanency of a subscription based model. It was a bummer that we couldn't export our maps in full resolution while we had our subscription in order to save them. We were forced to reluctantly let them go (like Rose at the end of the titanic movie.) For the maps we made with drone deploy, there was no way to preserve our work outside of the drone deploy cloud and we would have had to keep paying each year to avoid losing our results. It was a bummer (for us) that we couldn't have team members add annotations to our map without buying another entire full licensed seat for them. It's great that drone deploy is constantly improving their products. Maybe at some point in the future we'll have another project come through that can fund a year subscription and make use of their software. In terms of 'ease of use', there's really nothing easier. For our areas, they had better results and preserved better detail than pix4d. However, for our specific use case, our own in-house software actually gives us significantly better results than either drone deploy or pix4d ... everything takes time, but at some point I hope to put together a little blog post or a youtube movie or something to show the details of this more clearly. But again, our use case was finding needles in a haystack (not 3d modeling terrain, not doing ndvi on crops, not mapping construction sites) ... we are looking deep into the tiniest details of the images to try to spot things that aren't supposed to be there. Some times things that are supposed to be there spot us back ... !

 

[clolsonus]

Here's a one clear example of the sort of thing I am doing ... we are hunting oriental bittersweet which is an invasive vine with a distinctly colored berry. In the fall after the leaves drop (especially when there is a coating of snow on the ground the fruiting vines can become very clearly visible.) These infestations tend to be in forested areas where the vine can find things to climb and these are areas that are traditionally very difficult to stitch well. And even when successfully stitched, the resulting orthophotos can be a mishmash of parts of trees from different angles all munged and blended together. Those sorts of stitches can be really hard to make sense of because everything is completely disjointed. So here is an example of a single picture of oriental bittersweet ... it's a situation where you can't even see it in the top down photo, just the adjacent photos. (So a perfect orthophoto stitch would hide these vines beneath the tree's canopy.) The dense point mesh, the 3d model of the area ... all of those are cool things, but completely useless for seeing and geolocating our oriental bittersweet infestations:

 

[adm_geomatics]

Here's a one clear example of the sort of thing I am doing ... we are hunting oriental bittersweet which is an invasive vine with a distinctly colored berry. In the fall after the leaves drop (especially when there is a coating of snow on the ground the fruiting vines can become very clearly visible.) These infestations tend to be in forested areas where the vine can find things to climb and these are areas that are traditionally very difficult to stitch well. And even when successfully stitched, the resulting orthophotos can be a mishmash of parts of trees from different angles all munged and blended together. Those sorts of stitches can be really hard to make sense of because everything is completely disjointed. So here is an example of a single picture of oriental bittersweet ... it's a situation where you can't even see it in the top down photo, just the adjacent photos. (So a perfect orthophoto stitch would hide these vines beneath the tree's canopy.) The dense point mesh, the 3d model of the area ... all of those are cool things, but completely useless for seeing and geolocating our oriental bittersweet infestations:
View attachment 2077

Thanks for sharing. I can see how the state of the software 2 years ago wouldn't have been good in this use-case. Definitely allot different now.

 

[clolsonus]

Thanks for sharing. I can see how the state of the software 2 years ago wouldn't have been good in this use-case. Definitely allot different now.

In what ways has the software changed relative to my use case? Now that my license expired I don't really have a good way to know what's going on over behind the scenes at DD.

 

[adm_geomatics]

In what ways has the software changed relative to my use case? Now that my license expired I don't really have a good way to know what's going on over behind the scenes at DD.

There is now,

• Decreased allowable flight altitude from 70ft to 30ft
• Çrosshatch flight pattern with oblique pitch (this may have been in the beta app phase at that point, but it is now an integrated flight mode)
• Progress Report plan for repeatable static image positions
• 360 Panos
• Videos
• In flight pausing and manual capture.
• Point clouds are 20% more sense than 1 year ago
• 3D annotations ?
• Issues/Inspection Tool

I'm not going to disclose the details of our account in order not to upset anyone if we are different, but sharing with editors should not have been an issue..

 

[clolsonus]

Just a quick follow up (and not trying to pick on any specific commercial tools, just sharing my own personal experience ...)

Here is what pix4d did with one of my data sets (first picture) versus what I was able to do with my own tool chain (second picture.) I don't have access to see what drone deploy would do with the images any more due to our license expiring and not having more funding to renew.

Because of the significant elevation changes and needing to balance area and overlap, I had too much overlap in the lower areas and not enough overlap in the higher areas (where pix4d blew up.) But if you've driven 3 hours to fly an area, and done the best you could to balance all the constraints to get your data set, it would be nice to make full use of it.

 

[clolsonus]

Sorry to keep posting on this thread ... two more compare/contrast images. The first area is a detailed area of oriental bittersweet in the portion of the pix4d map that successfully stitched. You can see the artifacts of the stitching/combining process and the loss of details. The second picture is our in-house tools which have placed the original picture in it's correct map location at it's full original resolution.



In house tools:

 

[adm_geomatics]

Sorry to keep posting on this thread ... two more compare/contrast images. The first area is a detailed area of oriental bittersweet in the portion of the pix4d map that successfully stitched. You can see the artifacts of the stitching/combining process and the loss of details. The second picture is our in-house tools which have placed the original picture in it's correct map location at it's full original resolution.

View attachment 2080

In house tools:

View attachment 2081

No worries, threads become what they become and obviously no one else is posting on 3-D annotations which kind of was my expectation.

You obviously have a lot more contrast and sharpening built into your processing software. All softwares have their own exposure corrections and one seems a little light and the other seems a little harsh. I do all of my own photo corrections through ACDSee and have specific filters set up for each drone.

PM me a link to download your photos and I will process them in our DroneDeploy account. Once with your raw photos and again with my post-processing.

 

[clolsonus]

Hi Chasco,

Thank you very much for your kind offer to process my data set through the drone deploy tool chain (since I am no longer able to do that.) I am truly interested to see how it compares. I realize I am doing my own project a disservice if I'm not using the best tools (or close the best anyway) for the job.

Referring to your previous comment about contrast/sharpening. I do equalize the brightness a small bit which has the side effect of making the details sharper, but I'm not doing 'sharpening' or other post processing enhancements to the images. Also notice that much of the fine tree branch structure is lost in the pix4d orthophoto, the differences are in pixel resolution and chopped up or approximated details as well as due to tweaking the image brightness histograms a bit.

Here is link to a .zip file on my google drive. It's a collection of images taken in a state park (with the park managers permission) so there shouldn't be any issues with sharing or privacy since it's not private property. It's 19Gb (1490 images) which is sort of a medium size data set for our project.

old-hay-creek-big-circle-20200113.zip

drive.google.com

Pix4d managed to stitch about 2/3 of the images without too many artifacts. Here's what my in-house tools did (also not perfect if you run the visualizer tool locally and zoom in and around, also my in-house tools don't edge blend things together like DD/pix4d does, but it does at least show you how it is overlaying all the original images in their exact right places to make the map.)



Pix4d result for comparison (and pre-comparison against your drone deploy results):

 

[adm_geomatics]

Got it. I will get this downloaded and processed over the weekend. I’d be interested to see what your Pix4D settings were. Hopefully you didn’t process with half-size images.

 

[clolsonus]

Got it. I will get this downloaded and processed over the weekend. I’d be interested to see what your Pix4D settings were. Hopefully you didn’t process with half-size images.

I just uploaded the images to the pix4d cloud and took the result (exactly like you would do with drone deploy.) Maybe there are tweaks you can do to make pix4d come out better, but to be fully honest, I'm a linux guy and have never actually installed the windows software locally because I don't have a convenient computer to run it on. Anyway, back in the day when I owned both drone deploy and pix4d, I figured it was a fair comparison to upload the image sets to both and see what they did with their default settings.