hn-88 / ocvwarp Goto Github PK

Hello,

Could a version for Windows be compiled that works as a command-line tool? So, instead of opening dialogs, it reads three paths as command-line parameters.

Thank You,
Maciej Mucha

If we want support for multiple still image files + position each of them individually + animate each of them individually, it may be easier to enter all the details using a ui?

Seem to be discontinuous with the pixels above. Check map file.

The tool is amazing, and now works way much better still does not allow to be used as a commandline tool for images for two reasons:

• it does not allow to run on sequences starting not from 0 (so starting with frame 600 is not possible), so you cannot run it in batches, or in case of fail, you have to run it again from 0
• when using a single picture input (so literally written number, like: "Frame_00569.png" it opens a dialog with a question if it is a sequence or not. It should be possible to force him do the image which is not a sequence.

Submitter: HN
Testing gitreports.com

We could possibly add some parameters to ini file like
fps
number of frames
which would decide how long the video needs to be, instead of outputting just a single frame.

Of course, this can be worked around using avisynth, so this is just an enhancement.

Check out various ways of choosing your own codec. Adding a fourcc line to the ini file should be easy.

Currently anglex and angley - the viewing angles inside the 3d scene - can be varied on the fly using keyboard shortcuts. It would be easier to use if we can give initial anglex/y and final anglex/y, with a spline to ramp up rotation ... or at least constant velocity rotation.

Appveyor build at
https://ci.appveyor.com/project/hn-88/ocvwarp/build/artifacts
does not run on windows. Probably linked to some other dlls.

With Google Cloud Shell, got an image upside down. Will check if the same parameters give the same output on Windows.

As seen with hn-88/OCVvid2fulldome#10 - OpenCV's undocumented support for image sequences has the quirk that it assumes the input fps (based on locale?) - to be 25 fps (could be different for people in different locales). So, if processing a 30 fps image sequence, I would initially get a 25 fps video which I would then need to convert to 30 fps, for example using the method at https://hnsws.blogspot.com/2024/02/60-fps-to-30-fps-with-ffmpeg-without.html

Maybe can fix using the technique at https://dev.to/katkelly/changing-your-repo-s-language-in-github-5gjo

Submitter: HN2
More details

Don't work properly for non-zero anglex angley

This is the body of the message



via Email

January 16, 2021 at 09:08PM

For transformtype 4 and 5, multiple flips, remaps and resizes slow down processing. Explore the possibility of refactoring code to have a single remap, to speed up processing a little bit.

transformtype 1 goes at 10 fps while 5 goes at 3.8 fps. For the same input and output resolutions.

Currently, single frame input seems to work OK on Linux, to be tested on Windows. Output filename like output.mp4 works, from which the single frame can be extracted with avidemux.

https://stackoverflow.com/questions/73108273/building-and-distributing-a-macos-application-using-cmake
indicates macdeployqt

Bottom ten pixels are black. This seems to be due to clipping in transformtype=1 When transformtype=1 mapping is changed to allow r<=1.01, the extra pixels are not continuous.

One pixel wide seam (for 1080x1080 output) is seen for transformtype 0, 1, 5 where the leftmost image pixels meet the right-most image pixels.

For example, along centre of image when Anglex = 90, Angley = -90

Try an appveyor build. Check Windows build using a VM. If it doesn't work, try another build just removing the tinyfiledialogs. Because ViewportSaverc appveyor build works.

Try and create an appveyor build with mingw

for transform type 4 & 5.