A single image file should be able to be loaded so that it could go straight to the sharpening step

There seems to be a memory leak in some environments (ie. AlmaLinux 8, which is used to build the linux binary). It seems to originate from the pixbuf for the image preview and easily replicated by rapidly scolling the frame slider and watching the memory usage rise. Maybe the pixbuf needs to be manually cleared everytime?

Loading videos is single processed right now, it would be a speed gain to load using the specified number of processes.

Add frame normalization so that videos with varying brightness may have better alignment results.

The stacking step only uses a single process right now. It should use the specified number of processes.

Right now the images are automatically cropped based on the shifts. But it would be nice to not have it auto-crop and instead use the reference frame as a background. Drift shifts will still probably need to crop for now, but that might be able to be changed in the future as well.

The user should be able to specify a rectangle processing area so that the alignment only looks at that area to determine the transformation matrices. The actual transformation still occurs over the full image. And then the image similarity will also be only over the processing area

Having users need to set up a dev environment in order to run the application is not ideal, so there needs to be an installer, or single executable to run.

This has already sort of been started, it mostly already works, but there are a couple UI things which need to be removed for this to be fully functional.

Too much sharpening can cause ringing around stars so masking off sharpening in those areas should help.

Ideally the cache and frames folders should be changed maybe to some virtual filesystem or something like that. Also shouldn't have files like stacked.png and sharpened.png hanging around, those can probably just reside in memory.

Right now the UI doesn't do much in terms of validating inputs, and making sure that buttons are pushed at the right times. It should disable/enable fields depending on what step the user is on so that the program doesn't get into any fragile state.

First of all, this project is great - exactly what I was looking for as an open source alternative to AutoStakkert / Registax.

There's a couple of things that I think could be added in to make it more powerful, starting with an option to perform a deconvolution on the result image to address image quality issues in the telescope optics, such as imperfect focus, diffraction artifacts, or even motion blur from over-long exposures of individual frames.

OpenCV provides a pretty stright-forward example script for using cv.idft to preform a Wiener deconvolution. The simplest implementation would have the user provide an assumed PSF, likely in the form of a radius for a gaussian blur. A (much) fancier approach, which would be totally fair to declare out of scope for this project, would have the user photograph a star nearby to the planet of interest and use that image as the source for the PSF. They'd probably need to use an image stack there as well to avoid the computed PSF being mostly noise.

If you think this would be a useful addition, I may have some time to provide some help with implementation.

Having to read/write tmp files repeatedly adds quite a bit of overhead, and unnecessary code. No tmp files should need to be written.

Pillow is only used for the image similarity, and there shouldn't be any reason why opencv could just be used instead. Its probably increasing the size of the build.

Only 3 are enabled now, 5 will probably be a good amount to enable as long as performance holds up.

Will need to add a save dialog to save the final image.

The new file filter settings to include .SER files seems to be broken on windows

Some programs do not include an index in the videos which can drastically slow down the program execution. A solution might be to do something like VLC does and prompt the user to build the index (this would rewrite the video file). For now you can manually build the index by running the following:

ffmpeg -i input.avi -map 0 -c copy output.avi

It would be nice to let the user know that a new version has been released. Can probably use this api:
https://api.github.com/repos/Finalfantasykid/AstraStack/releases/latest

Multiple images should be able to be selected so that they could be the frames being stacked instead of video frames

When during the alignment step it would be nice to be able to specify a custom start/end frame. The drift adjustment could use these as well to produce a more accurate drift adjustment.

There should be a warning letting the user know that they are going to overwrite an existing file, and give them a chance to cancel.