OpenStreetMap logo OpenStreetMap

Post When Comment
RoboSat ❤️ Tanzania

Sure Jesus.

I believe the problem was images and labels are not in sync. Try run the following commands.

$find project/zanzibar/dataset/ -type f cut -d/ -f-5 uniq -c

where change ‘project/zanzibar/dataset/’ into your path where ‘dataset’ folder is located.

Mine return the following.

  1     project/zanzibar/dataset/evaluation/images   43080 project/zanzibar/dataset/evaluation/labels
  6     project/zanzibar/dataset/training/images  344584project/zanzibar/dataset/training/labels
  1     project/zanzibar/dataset/validation/images   43074 project/zanzibar/dataset/validation/labels

Obviously, my images and labels are not in sync. The problem i found is that, I have the correct building.tiles which I want to focus to, but I couldn’t download or find the correct GEOTIFFS for the respective building.tiles ( i believe so). I wanted to find the area of Putrajaya, Malaysia.

So, I’m stuck doing this for the last 3 days. I’m clueless.

Thank you, Fairuz,

RoboSat ❤️ Tanzania

Hi Jesus Benavent,

The file dataset-building.toml is already there under directory,

robosat-master/config/dataset-building.toml

You can edit the file accordingly.

Fairuz,

RoboSat ❤️ Tanzania

Hi Daniel,

I have another questions. I got an error while running the ‘rs training’. Below are the errors.

File “/robosat-master/robosat/datasets.py”, line 58, in init assert len(self.target) == len(self.inputs[0]), “same number of tiles in images and label” AssertionError: same number of tiles in images and label

What does it tell?

Thanks,

RoboSat ❤️ Tanzania

Hi Daniel,

I got the answer already on how to split the building.tiles. Basically, we need to split this building.tiles files into 3 (training, validation,evaluation) using any linux script for example sed, awk in bash script.

thanks.

RoboSat ❤️ Tanzania

Hi Daniel,

Can you explain how to perform training split on 80:10:10 ? Do I have to make changes on the files building.tiles or validation.tiles?

Thanks in advance.