DJI Mavic 3M - Dataset

Hello Guys!

I’m creating this topic because I was trying to reply a post from @AlfredWang (DJI Mavic 3M) but, unfortunately, it was closed.

I’m looking for multispectral pictures from DJI latest Mavic 3M but i couldn’t find any pack yet. If someone have similiar images and will to with me I would be so grateful.
If someone have a similiar pack of pictures or can communicate with him, I would be so grateful :slight_smile:

If AlfredWang see this post, please leave an email where I can communicate with you

Thanks!

2 Likes

Hi livia,

Sure, I can share my datasets. I’ll upload it to my google drive.

I have two datasets, they have different flight altitudes: 150 meters and 80 meters. The 150 meters one is about 3GB, another one is 14GB. So it’ll take some time to upload, when it finished, I’ll let you know and share the link.

you can contact me with:
[email protected]

cheers

1 Like

Hey Alfred,

thank you for answering! I’ll contact you by email now.

Cheers!

1 Like

Hi guys, I’ll put links of DJI Mavic 3M datasets here, anyone can access and use.
There are two flights with 150 and 80meters altitudes.

If you are doing agriculture RS and need more info about the field, feel free to contact with me.

150 meters:
https://drive.google.com/drive/folders/1tHmy5497IXjWBCbPbTomoZclIAxG6M1O?usp=share_link

80 meters:
https://drive.google.com/drive/folders/1HQwfkoSCPRzBLqjfNqQPumEHOInxPoBy?usp=share_link

3 Likes

Hi Alfred,

Thanks for the MS images. I took a quick look at your datasets and found something that (probably) results in the error you described in DJI Mavic 3M

When I looked at your 150m dataset I saw that some images are missing. For the MS images at each GPS location 4 tiff’s are taken (…G.tiff , …NIR.tiff , …R.tiff and …RE.tiff)

When you sort the images by (filetype and name) you can see that some images are missing per group. The red brackets/groups are incomplete.

My guess, when you remove all the “red” images WebODM can proces them correctly.

The initial dataset was 186 images (with 4 bands) → 186 images / 4 bands = 46.5 “groups”. A round number is expected, this triggered my into looking if all the tiff images were stored.

The dataset with the incomplete image-groups were removed, 180 images / 4 bands = 45 groups. This is a round number and that should work (processing is currently running).

I guess Pix4D can “skip” an incomplete group of MS images where WebODM results in a processing error.

Funny is that I had the similar issue with the Phantom 4 MS, not all tiff images were stored or were 0kb. When I removed these incomplete groups it worked like a charm.

2 Likes

Hi Alfred,

150m Result seems ok. Processed with WebODM and ODM version 3.0.2

2 Likes

Hi Aeret,

Thanks for your reply.
I checked the dataset and make sure groups for all bands are complete. Then I tried again and get some good results after upgrading my RAM to 40GB (from 16GB). So, I’m wondering the problem I met is caused by running out of memory. But interestingly, the error messages are not always like “lack of memory”, sometimes it’s just crashed.

Here are some results in QGIS:


The average GSD is about 7cm for 150 meters dataset.

2 Likes

Silly question, here, but just trying to work out multispectral. Is it just the images from the multispectral camera you upload from a single folder? ie do you leave out the RGB imagery?

2 Likes

Yes, I filtered all images by filename extensions and selected multispectral images(.tif) into a single folder, as you mentioned.

I never try to mix .jpg images with .tif images together and process, I’m not sure if it can work.

1 Like

Thank you!

2 Likes

We used to not be able to handle this, but if you’re on the current/latest release it should handle this gracefully.

We still however, need multi-band datasets to have the same number of images per each band to work properly.

2 Likes

COOL!

Is the final result two orthoimages? The one is higher GSD with RGB bands, and another is lower GSD with MS bands?
If yes, that’s awesome and really simplify my working flow.
If not, how does it manage the data gap between RGB and MS images? As I know, the RGB camera usually has much higher resolution than MS cameras but lower pixel depth.

1 Like

If my understanding is correct, we currently prefer the multispectral data and use that for the entire reconstruction workflow.

1 Like