Currently outputs are created using the built-in thumbnail function from the image crate. This is great to get things up and running, but in cases where the input file is large we should instead compute the the output image directly.
Since we know the input size and the desired output, we can define operations to map relevant input data to each output pixel.
This is done for SICD data, need to make things more generic or copy-pasta for other image types.
The brunt of this work is just going to be sorting out the array shape required to efficiently iterate over the image. As an initial pass, only implement non-blocked and 'simply blocked' (I don't remember the NITF parameter combination for this, but most data seems to be written like this)
holmesv3
commented
6 days ago
Currently outputs are created using the built-in
thumbnailfunction from theimagecrate. This is great to get things up and running, but in cases where the input file is large we should instead compute the the output image directly.Since we know the input size and the desired output, we can define operations to map relevant input data to each output pixel.
This is done for SICD data, need to make things more generic or copy-pasta for other image types.
The brunt of this work is just going to be sorting out the array shape required to efficiently iterate over the image. As an initial pass, only implement non-blocked and 'simply blocked' (I don't remember the NITF parameter combination for this, but most data seems to be written like this)