Recovering video from 720 x 1280 drone cache

Hello! I googled everything I could on this topic, but I haven’t found a solution yet. I have this problem. I filmed very good content on a trip to the Philippines, but the drone fell into the ocean and drowned and I didn’t have time to back up the flash drive(
There is a cache left - not only is the video in 720, but the details are also very compressed. I tried Topaz Video AI - but unfortunately it does not improve essentially anything - the same thing, even a little worse, but the file is 4K in resolution.
I think there might be a problem with the settings. I have one of the latest versions 4.2.1.
The first thing I tried was to set 4K, 30 or 60 frames per second Progressive with the AI model Proteus and the default Recover Detail 20 (tried different values up to 100)

  • I tried to set the parameters manually - but I don’t know how to do it correctly
    As a result, everything is just as bad with this mode, only the video is in 4K
    Iris, as I understand it, is for faces - I have landscapes
    Nyx for noise - I have everything there in bright sunshine, no need to adjust noise
    Artemis with the low quality option - comes out a little better than Proteus but not much
    Gaia - as I understand it for the computer. graphics are more suitable, but I tried it too - like Artemis it comes out a little better than Proteus but not much
    Theia - I don’t quite understand why it is needed, I haven’t tried it…

P.S. trying all this is quite time-consuming - even 10 seconds takes a long time to process.

Maybe you can advise which mode to choose from this whole variety in order to get a good result, perhaps there is the same Proteus with special manual settings?

In my video, the content of beautiful tropical views was compressed in low resolution with the loss of some details. There are no people in the video, everything is in good lighting…

Please upload the video file somewhere so that I can give it a try! Processing 4 k video takes some time. That is quite normal. :slight_smile:

1 Like

Cache is simply a copy of your FPV video feed. It is 720p.
However, it is not a downsampled copy of the drone’s video, for example, if RC signal becomes weak, your FPV (and cache) will be blocky, low res, even monochrome.

Put simply you’re not going to be upscale to 4K and think you will have the same quality as footage recorded by the drones camera.

1 Like

Please upload a sample to my dropbox or post a public link so we can test and make some recommendations

1 Like

Thank you! When i insert my public link, i have this message - “An error occurred: Sorry you cannot post a link to that host.”
That’s why I uploaded the files to your dropbox. There are two small files. I just tried to do what I described in this topic with these files

Thanks a lot. Managed to upload to dropbox margaux.topazlabs - Dropbox

1 Like

Yes, I understand that this is not just an increase in resolution, but restoration of details, which is possible with the neural networks that topaz video ai uses. I read about this, but in practice I didn’t see much of a difference - perhaps I’m using the wrong settings, as I wrote about in this thread above

I will have a look into it soon. Please be patient. :slight_smile:

1 Like

I am not from topazlabs. I am just alpha and beta tester. Can’t get access to the file there. :confused:

please - 2024_03_24_14_10_06_Cache.mp4 - Google Drive
and - 2024_03_23_13_47_37_Cache.mp4 - Google Drive

1 Like

Your source files were HD 30 fps (1280 x 720). I made four files, two are 1280 x 720 in 60 fps and another two are 4 k (3840 x 2160) in 60 fps via Apollo:

These are the settings I used:

1 Like

I didn’t put to much time and effort on this. I upscaled it to 1080p and I did not change the fps.
But I gave you a base line to start with that you can work with to tweak your settings.
all were processed with Prob4, two sets are at 50/70 (50=Fix Compression, 70=Improve details, everything else at “0”) and the other two sets are at 40/40.



Play around with the Fix Compression & Improve details until you get to your desired looks. Start with increments of 10 on one or both. add some and remove some. e.g. 30/50 , 50/60, 40/60, 20/40, 20/50, etc.

BTW- this was captured in El-Nido in Palawan?

1 Like

thank you very much, but it seems to me that now the quality has become worse than in the source code, unfortunately

Yes. I ramped up sharpen and recover details to full. Maybe lower the values for a more natural look.

1 Like

Thanks a lot! Some of these remastered ones are actually better than the original. 2024 03 23 13 47 37 Cache prob4_50-70.mp4 turned out well and 2024 03 24 14 10 06 Cache prob4_50-70.mp4. Thank you!

“BTW- this was captured in El-Nido in Palawan?” - Yeah almost. It was there nearby - Coron, Busuanga

These manual settings are exactly what I was looking for. But the meaning of some is not entirely clear.
For example -

  1. “Fix Compression” is probably the option that should correct problems and losses of the part, but then why 50, if unscrewing it 100% is bad?
  2. The same applies to the “Improve details” option, why not set it to 100% - what side problems might appear?
  3. The “recover details” option is even less clear - it restores details better, but it is often left at 20. Is it better to leave it this way?
  4. “anti-alias/deblur” is set to zero, but what other values give and where to move the slider is not at all clear.

Yes, perhaps the problem is that there are the highest sharpening values.
But I still don’t understand why Apollo v8
It’s also not clear why Iris is used - is it mainly for treating faces?

I used Iris V2 MQ because it feels quite creative in recovering details. Not only on faces. Apollo 8 tends to generate less artifacts than Chronos when interpolating and was recently suggested as one of the most capable interpolation models currently, that’s why I thought I give it a try. The “Improve Detail” option for my impression does not necessarily improve details but makes existing ones more visible. “Recover Details” on the other hand tries to keep a certain amount of the original image. This might lower the effects of the other options.

ohh yeah, I wasn’t sure. I was there too. did diving seeing WWII skunked ships in Busuanga

  1. the Fix compression should fix the blocks in the video you have, but to High will kill it’s natural looks

  2. to high over sharpens the footage and creates artificial looks as well. as you can see in my 50/70 it was sharp and clear but the cliffs didn’t look natural, with the 40/40 i did you see it is more natural with less sharpness. that is why you need to play around with those values to find your sweet spot for your Drone videos.

  3. Recover details basically what it means, how much of the original video footage you want to be backed in the processed file to contain and how much is AI.
    I never really played with this option or saw any significant difference in my videos that worth investigating deeper on that option, at lease for my content.

  4. it was never cleared to me either. no matter what Direction I move the slider, I get the same results of deblurring and not much difference on other parts of the video.

Another thing. the reason Why I did not attempt using Iris, is because Iris is more intended for faces and your videos are more scenery footage , so Proteus is much more adequate than Iris for this type of footage.