clarification on the custom dataset creation and inputs during the inferences
I’ve tried the new ibm-nasa Prithvi 100m model but during the inference or the testing stage I stumbled around a bunch of hurdles. I Experimented with Scar and Burn model on Hugging Face with their examples to run the inference which gave me correct outcome as expected with a masked region of Burn area, but the problem arises when tried using the training data for the inference the model seem to work and doesn’t provide any prediction or the masked region.
Maybe I am giving the inputs not correctly and doing something wrong.
Can someone in the community help me out with solving or throw some light on the dataset. My queries are listed
1.How should we create our own geotiff dataset for ibm-nasa Prithvi
a. Currently I am using the training dataset to give the model input .
I wanted to get some guidance for the custom dataset creation so that I will be able to test it for various test cases.
guidance for tif and masked files creation and the merging of 6 bands.
2. How to create a merged tiff file with all 6 bands
a. While going through the dataset I came across with specificity of the bands the dataset contains that are
1, Blue, B02
2, Green, B03
3, Red, B04
4, NIR, B8A
5, SW 1, B11
6, SW 2, B12.
Can someone provide some information on the usage of these bands in the dataset and how to provide these bands as the inputs
- One or two sources to get the data of satellite images
a. I went through some of the dataset resources such as https://hls.gsfc.nasa.gov/ and https://mtbs.gov/ which was cited in the Hugging face repository but
was not able to find any of the dataset suitable for Burn and Scars inference