Messages

1

General / Re: Mavic 3 Themal image processing

« on: February 22, 2024, 02:01:13 AM »

The trick is that DJI encodes the thermal info into the thermal images in a proprietary way.
When you view the images in a normal image viewer or in Metashape, you are seeing and RGB render based on whatever colourbar you selected during the flight.
The actual radiometric information underlying those colours is saved in those images, but Metashape doesn't extract it for you.

There are a few options available to you:
We used to use Thermoconverter (https://www.thermoconverter.com/). It did the job, but cost money and wasn't scriptable/automatable, which is something we wanted to do.
There are various other equivalent bits of software out there, but they all cost money also.

So instead I built up a custom Docker image (very loosely based on https://github.com/daz/dji-thermal-sdk-docker) but tailored for our use case in Google Cloud Platform.

Other things that might be worth exploring:
https://www.dji.com/au/downloads/softwares/dji-thermal-sdk
https://pypi.org/project/dji-thermal-sdk/

Once you have converted the images (by whichever means you like), Metashape can import those converted outputs and work with them to create a radiometric thermal mosaic just like normal.

2

General / Re: False Color ortho color issue - DJI M3M

« on: February 05, 2024, 05:12:19 AM »

Hi there,

I think the main problem is that what you have in your screen shot as bands 1-7 is made up of the non-radiometric RGB data as bands 1-3, and then the mutispectral bands as 4-7.
This can be seen because you have 2 x "Red" and 2 x "Green" in  your list.
If you tried the same raster calculator process using only the 4-7 numbered bands, it might work better.

More generally, we tend to process the MS in an entirely separate project from the RGB photos. Maybe worth considering.

Cheers,
- Mike

3

Python and Java API / Running Metashape on a headless VM means no multithreading

« on: October 18, 2023, 09:28:55 AM »

Hi there,

We're using Metashape as part of a fully automated pipeline, so it is running on a virtual machine (VM) in Google Cloud.
We have it running succesfully completely headless after a bit of fiddling around with the display settings.

Currently, to get it running we have the following lines in a bash script (in amongst some other stuff)

Xvfb :1 &
PID=$!
DISPLAY=:1 /usr/local/metashape-pro/metashape.sh -r /usr/local/metashape_automation_repo/metashape_automation.py
kill $PID

Since adding this we have noticed that the software is no longer using all the CPU available to it. It still seems to be using the GPU, however.

My linux/bash isn't good enough to know if this is the cause of the problem, but if there are any linux gurus here I'd love to hear if this is the only way to achieve headless operation?

But I'm asking more generally, also, if there are any flags or settings that can force Metashape to be multithreaded, or specify the number of CPU  to working on.

Or can we get it to print how many CPU it sees, so we can check if it is indeed only seeing a single CPU?

4

General / Re: generating a multispectral orthophoto with Mavic 3 multispectral

« on: August 07, 2023, 09:07:02 AM »

Hi there,

I just thought I'd post that we are processing our M3M data in Metashape just fine.

Key steps:
- Add Photos, and select only the .tif files
- Load as a multicamera system. This will appear that only the _G files have loaded (green band), but I promise they are all in there
- Run alignment
- We treat the point cloud densification as optional - we don't care about the DSM, and skipping densification saves heaps of time
- Create DSM
- Calibrate Reflectance (use spectral panels if you have them, otherwise just tick the box for the sun sensor)
- Mosaic with "average" blending - we get stripes in the mosaic if we use the other options. I've never found an official reason for it, but my guess is that its due to the pitch of the drone as it points towards and away from the sun with North-South transects
- Don't forget to convert to reflectance afterwards (we do this in QGIS as part of our QAQC process)

Hopefully that helps someone!