Can you use srestore to remove blends then rife to rebuild removed frames?

[LaserBones]

I've been working on encoding an old public access show I recently acquired DVDs for. I've had to approach every episode separately since they all have their own problems. Some of the episodes have had every other frame blended. Thanks to Jagabo I've used rife to fix these videos by selecting the even frames and rebuilding the missing frames with rife.

So here is my current problem. For whatever reason on this episode the blends that are normally every other frame randomly switch every couple minutes. So odds become blended instead of evens and a few minutes later the opposite happens. Because of this I can't just discard even frames and use rife. Is there a way to use restore or something similar to remove blended frame and then somehow rebuild only the removed frames regardless if they're on a consistent cadence?

[LaserBones]

Made some progress. This is working for the most part. Still a few blends. Is there a better way to do this?

srestore(omode=1)

Selecteven()

z_ConvertFormat(pixel_type="RGBPS", colorspace_op="709:709:709:l=>rgb:709:709:f")
Rife(gpu_thread=1, model=9, fps_num=30000, fps_den=1001, sc=true)
z_ConvertFormat(pixel_type="YUV420P8", colorspace_op="rgb:709:709:f=>709:709:709:l")
MCTemporalDenoise(settings="low")

[poisondeathray]

Made some progress. This is working for the most part. Still a few blends. Is there a better way to do this?

srestore(omode=1)

Selecteven()

z_ConvertFormat(pixel_type="RGBPS", colorspace_op="709:709:709:l=>rgb:709:709:f")
Rife(gpu_thread=1, model=9, fps_num=30000, fps_den=1001, sc=true)
z_ConvertFormat(pixel_type="YUV420P8", colorspace_op="rgb:709:709:f=>709:709:709:l")
MCTemporalDenoise(settings="low")

A better way would be to specify the correct framerate in srestore, instead of using selecteven

srestore(frate=something)

[LaserBones]

Made some progress. This is working for the most part. Still a few blends. Is there a better way to do this?

srestore(omode=1)

Selecteven()

z_ConvertFormat(pixel_type="RGBPS", colorspace_op="709:709:709:l=>rgb:709:709:f")
Rife(gpu_thread=1, model=9, fps_num=30000, fps_den=1001, sc=true)
z_ConvertFormat(pixel_type="YUV420P8", colorspace_op="rgb:709:709:f=>709:709:709:l")
MCTemporalDenoise(settings="low")

A better way would be to specify the correct framerate in srestore, instead of using selecteven

srestore(frate=something)

Sorry, I may have not been clear in my first post. The original video from the DVD has hardcoded blends in every other frame. So I'm purposefully dumping those. The other problem is the blending randomly changes depending on the scene. Sometimes it's blend, frame, blend, frame. Other times its frame, blend, frame, blend. Using omode=1 is getting rid of most of the blends but when it does flip cadence I do get one blend at the start of that scene with the current script I'm using.

[poisondeathray]

Sorry, I may have not been clear in my first post. The original video from the DVD has hardcoded blends in every other frame. So I'm purposefully dumping those. The other problem is the blending randomly changes depending on the scene. Sometimes it's blend, frame, blend, frame. Other times its frame, blend, frame, blend. Using omode=1 is getting rid of most of the blends but when it does flip cadence I do get one blend at the start of that scene with the current script I'm using.

Right, the pattern changes - so I'm thinking it's probably better to let srestore get rid of the blends by looking at all of the frames , instead of relying on selecteven . Srestore is adaptive. SelectEven is not

The way are using it now, srestore in omode=1 duplicates a detected blend with a previous frame. But what if it's right on a scene change - eg. the pattern on the next scene starts with a blended frame - that might be causing the remaining blends depending if that frame number was odd or even

Instead, if you use srestore in omode=6 with frate=14.985 (because 29.97 / 2, assuming this is NTSC) look at all of the frames and choose. It decimates the blends (hopefully) leaving you the unique clean frames. Then use RIFE . I don't know if it will be better on your specific source, it's just a guess based on your description

Sometimes using a slower speed can increase the accuracy eg. speed=-25 . Also make sure you check a linear encode, or advance a few seconds back if checking (sometimes srestore can mix up frames if you seek non linearly)

[LaserBones]

Sorry, I may have not been clear in my first post. The original video from the DVD has hardcoded blends in every other frame. So I'm purposefully dumping those. The other problem is the blending randomly changes depending on the scene. Sometimes it's blend, frame, blend, frame. Other times its frame, blend, frame, blend. Using omode=1 is getting rid of most of the blends but when it does flip cadence I do get one blend at the start of that scene with the current script I'm using.

Right, the pattern changes - so I'm thinking it's probably better to let srestore get rid of the blends by looking at all of the frames , instead of relying on selecteven . Srestore is adaptive. SelectEven is not

The way are using it now, srestore in omode=1 duplicates a detected blend with a previous frame. But what if it's right on a scene change - eg. the pattern on the next scene starts with a blended frame - that might be causing the remaining blends depending if that frame number was odd or even

Instead, if you use srestore in omode=6 with frate=14.985 (because 29.97 / 2, assuming this is NTSC) look at all of the frames and choose. It decimates the blends (hopefully) leaving you the unique clean frames. Then use RIFE . I don't know if it will be better on your specific source, it's just a guess based on your description

Sometimes using a slower speed can increase the accuracy eg. speed=-25 . Also make sure you check a linear encode, or advance a few seconds back if checking (sometimes srestore can mix up frames if you seek non linearly)

Wow, that makes perfect sense. Thank you!

[LaserBones]

[QUOTE=LaserBones;2718263][QUOTE=poisondeathray;2718262]

Instead, if you use srestore in omode=6 with frate=14.985 (because 29.97 / 2, assuming this is NTSC) look at all of the frames and choose. It decimates the blends (hopefully) leaving you the unique clean frames. Then use RIFE . I don't know if it will be better on your specific source, it's just a guess based on your description

Wow, that makes perfect sense. Thank you!

I just realized there are a few segments where there are no blends (29.97fps). Do you have any advice on what to do there since there are no blends there? Should I still just set frate to 14.985 and let restore try to sort it out or is there a way to preserve those frames? Ultimately I want to get rid of the blends since they make the footage unwatchable so don't mind sacrificing the frames if needed.

[davexnet]

[QUOTE=LaserBones;2718265][QUOTE=LaserBones;2718263]

Instead, if you use srestore in omode=6 with frate=14.985 (because 29.97 / 2, assuming this is NTSC) look at all of the frames and choose. It decimates the blends (hopefully) leaving you the unique clean frames. Then use RIFE . I don't know if it will be better on your specific source, it's just a guess based on your description

Wow, that makes perfect sense. Thank you!

I just realized there are a few segments where there are no blends (29.97fps). Do you have any advice on what to do there since there are no blends there? Should I still just set frate to 14.985 and let restore try to sort it out or is there a way to preserve those frames? Ultimately I want to get rid of the blends since they make the footage unwatchable so don't mind sacrificing the frames if needed.

Perhaps handle the sections separately, bring them to a common fps and then join them