How to make own time-series dataset for semantic segmentation？ How do you generate datasets that can be used for training from satellite images and corresponding labels

Hi @VSainteuf , how is it going?

I have been studying the code related in terms of training with different datasets. And one question that I have faced based on the data that you have used is about the labels created into semantic $TARGET_{}.npy$ and $ParcelIDs_{}.npy$. I have attached a figure of patch 10006:

As you can see, in instance annotation before mapping with semantic channel 0 ($ParcelIDs_{}.npy$), we have a set IDs that isn't related with semantic annotations (in terms of background class, specially). Besides of that, in channel 2 of $TARGET_{}.npy$ we have a set of regions, related with background class, that isn't labelled as in channels 0 and 1.

So, I would like to better understand what was the methodology applied in labelling instances ($ParcelIDs_{}.npy$) and semantic label at channel 2 ($TARGET_{}.npy$).

Thanks a lot!

I am struggling with very low/non-existent GPU usage when trying out PSETAE on the pixel-set dataset. Because of this, I am seeing if this can be fixed by using the cache=True option to load the dataset into memory. Unfortunately this is taking a long time, >1 hour!

Please may I check a) how good the GPU utilisation can be without preloading and b) what reasonable times one can expect it to take to load into memory?

Thanks

Sam

Hello. I want to use your dataset in my project, but first I need to know the order of the spectral bands in your data. Is the following the correct order of the 10 bands (channels) in your data? Blue, Green, Red, Red Edge 1, Red Edge 2, Red Edge 3, NIR, Red Edge 4, SWIR 1, SWIR 2

Hi, I want only to select the particular date of images rather than a series. By setting attribute mono_date = "2018-09-02" on PASTIS_Dataset(). It raises

Unsupported operand types(s) for -: 'dict' and 'int' error

Hi, May I know how to visualize multi-spectral image (cube)? The code shows to visualize RGB images, however, if I want to show multi-band, how can I do that? Thanks in advance.

Dear Vivien, Thank you so much for your generous sharing pastisR datasets, However, when I try to load the dataset, I found maybe there is a bug of the dataloader.py about self.mono_date. Beacause in the demo.ipynb, the default mono_date = None, thus there will be:

     89 self.mono_date = (
     90         datetime(*map(int, mono_date.split("-")))
---> 91         if "-" in mono_date
     92         else int(mono_date))

TypeError: argument of type 'NoneType' is not iterable

I think it's caused by your correction in #14, To fix this problem, I make a simple alteration：

if mono_date != None:
            self.mono_date = (
                datetime(*map(int, mono_date.split("-")))
                if "-" in mono_date
                else int(mono_date))
else:
            self.mono_date = None

Then it worked, my idea is not necessarily correct, just for your reference. Thank you for your sharing again! ^_^

PASTIS_Dataset seems to require a metadata.geojson file, however it doesn't seem to be anywhere in the repository or available for download on the zenodo page.

https://github.com/VSainteuf/pastis-benchmark/blob/5710fe2933d21c035970b40ff261516ed760c1dd/code/dataloader.py#L105

Could you please upload this file?

Hi, I have a few questions on the Sentinel 2 bands. I would like to know in which order are they stored. Looking at the input bands, it looks like they are stored in order of wavelength. However, I have some doubts on the Sentinel 2 band 8 (stored at index 6, I think). Indeed, this band from Sentinel 2 is at 10 m resolution. However, on some patch time series, the band 8 has an odd aspect. I have displayed two examples : This example is a “normal” patch time series. The band at the index 6 appears at 10m resolution. In this image (each column is a different band)

In the second image, the band at index 6 has an odd aspect, like it was over sampled by a KNN :

I was wondering what could cause this effect. Best regards,

Hi,

I was just wondering what the license of this code is. I didn't find any LICENSE.md or similar in the project.

Thanks for this wonderful benchmark dataset!

Hi Thanks for sharing your code. Very nice code and paper. I read your paper carefully and analyzed your code. There are some files which I dont understand in the dataset. In your data loader function you are loading four different types of npy files for panoptic segmentation. Among them some files are named as ZONES_id.npy, what are these files are for. When I randomly checked one of them I found that they are very similar to instance_id.npy files. I think if you can explain a little bit more of your dataset and what data files represents what, that would be easier for serious reader and researches.

kind regards