itch.io

Update: So, the manual splitting doesn't help after all. I found the artifacts even in small segmented videos. The most typical case is the blue sky. So it seems I cannot use the DAIN app on these videos. :-(

Update: So, after some testing I found a solution. Changing the section size or padding doesn't help. The only workaround is to split the input video before actual processing. I use a combination of batch script and ffmpeg so this is not a big deal but I then have to run the DAIN process on all segments (25 segments in my case) manually.

Is there any way to control the DAIN app via the command line so that I could automatize the whole process? It would be extremely helpful for me!

Update: I tried reducing the video size to 624x624 pixels and the artifacts are not present so it seems to me that a) there is a bug in the splitting process, or b) the video segments needs more padding. Will do further tests...

Hello, I am using DAIN APP 0.31 on fulldome videos (special format of video captured with a fisheye lens and used in digital planetariums). The algorithm performs great in terms of motion smoothness but sometimes, it produces strange artifacts. They typically occur in large homogeneous areas such as a blue sky (rarely, they occur elsewhere). Here's a sample:

Thank you. Still, it would be much easier for users to be able to read the info on the main page, without digging deep in obscure chat servers. ;-)

I would be very grateful if the devlog could be kept up-to-date. I am now testing the app on fulldome videos with 4K+ resolution so I very much appreciate the version 0.3 which has introduced the split option. However, it is hard for me to keep track of development. Thanks very much for the effort anyway, I will surely spread the word among the fulldome community!

Similar behaviour here. But if you click on GPU to view the details you can see some activity in the memory copy section. My guess is that the algorithm itself uses a lot of memory accesses but not so much computations. That's generally not ideal for GPU acceleration. However, the application seems to be some kind of straightforward build from the python code so the GPU code most likely comes from python libraries. So there definitely is a room for optimization but it would probably require rewriting the app from scratch in CUDA C/C++.

AMD supports OpenCV which is a different programming framework. So, it would require completely rewriting the application which is not that easy I guess. :-)

From the screenshot you seem to have a similar problem as me. Almost 6 GB of memory is already allocated by other applications. Try closing all the programs (especially video/image editors such as Photoshop, Premiere etc.). Also try to connect your monitor to an integrated GPU if possible. That way, you'll have all the 2080's memory available for DAIN.