Posts: 393
Threads: 23
Joined: 2019 Oct
Thanks: 385
Given 526 thank(s) in 185 post(s)
Country:
Second render looks better. Comparing the 2K mxf to the Resolve Rendered 2K H264 the picture does look sharper. I'm also noticing that the colours of the mxf are looking nicer; darker but overall more defined, if that makes sense. I'm finding the blacks are much darker too, but not in a crushed way like when I set the video levels wrong.
Obviously these screenshots will have additional compression etc but they at least show a bit of the difference in detail and colour.
H264 - MP4
DNxHR - MXF
And for reference here's the "crushed" version where I set the data levels wrong.
I'll see what kind of result DNxHR gives me with a 4K render and report back.
I use Handbrake a lot for encoding DVDs. I've heard it uses x264 for it's encoder, would you recommend using it for rendering MXF files to Hx264 MP4/Mov?
Program material is recorded on the other side of this disc...
deleted user
Unregistered
Thanks:
Given thank(s) in post(s)
Strange, there shouldn't really be any difference in color at all. Which one would you say looks closer to how your timeline looks? Maybe one of the encoders is a bit messy. Could also try DNxHD in a Quicktime container.
Yeah I think Handbrake should be fine if you set the parameters right. x264 quality is all about setting the parameters or presets right. So for example the preset Slower should be a baseline I think.
Posts: 393
Threads: 23
Joined: 2019 Oct
Thanks: 385
Given 526 thank(s) in 185 post(s)
Country:
I'd say the H264 - MP4 more closely matches the colours I see in my timeline. Could it be something to do with Resolve and VLC interpreting 12 bit colour differently? I think by default Resolve's timeline is set to 10bit, the H264 - MP4 was rendered at 8bit, and the DNxHR - MXF is still 12 bit.
The colours of the MXF do seem to "pop" more, but I can't work out if I'm getting a bit of black crush.
It'll be another hour and a half before the 4K test finishes rendering, but once that's done I'll try rendering the MXF files to H264 using Handbreak.
Program material is recorded on the other side of this disc...
deleted user
Unregistered
Thanks:
Given thank(s) in post(s)
It could be a VLC issue, not sure. Try another player maybe. But if you're consistently getting such differences, I would not use this method as it makes it pretty unreliable. You're supposed to get exactly what you see in Resolve after all.
Posts: 393
Threads: 23
Joined: 2019 Oct
Thanks: 385
Given 526 thank(s) in 185 post(s)
Country:
I'll try another player once the video has finished rendering. If I do anything else while my PC renders 4K it just fails.
Agreed, I want to be getting the same colours. At the very least I'm seeing the difference in detail levels between Resolve's 2K H264 and DNxHR implementations, so if nothing else this should help answer the question of whether there's really enough detail to go 4K.
Once again thanks for all your advice and patience
Program material is recorded on the other side of this disc...
Posts: 393
Threads: 23
Joined: 2019 Oct
Thanks: 385
Given 526 thank(s) in 185 post(s)
Country:
2020-05-31, 11:44 PM
(This post was last modified: 2020-05-31, 11:49 PM by HippieDalek.)
I just rendered the 2K DNxHR MXF to x264 MP4, and to my eyes the output is identical to Resolve's:
Resolve's Render
Handbrake's Render
On the plus side the colours are back to being the same as they were in the timeline, so that's probably just a quirk of playing the 12bit MXF file.
I set the "Slower" Encoder preset but let the rest of the settings a default. Tomorrow I'll have a read of some of the encoding threads here and see if I can tune the settings a bit better.
EDIT: Just realised I forgot to check Constant Framerate. Schoolboy error.
Program material is recorded on the other side of this disc...
deleted user
Unregistered
Thanks:
Given thank(s) in post(s)
I'd say there's a bit of an improvement, most noticeable if you look in the bottom left corner, under the table. But generally you would want to compare a few different scenes, particularly some with a lot of movement. And the scene you picked is one of those that seemed to fare the best out of all anyway. In a lot of scenes in the Grindhouse release the grain is barely even noticeable, particularly in dark areas. That's usually where x264 shines due to its aq functionality which gives more bits to dark and flat areas.
Edit: By the way, you wanna set Constant Framerate I think.
Posts: 393
Threads: 23
Joined: 2019 Oct
Thanks: 385
Given 526 thank(s) in 185 post(s)
Country:
Ha ha, yeah I noticed the framerate setting as soon as I posted the picture, real schoolboy error.
Yes, I see the differences your meaning now, well spotted. I've picked this scene as I think it's a good one to start the reconstruction with; it displays most of the problems that I'll have to deal with including; dirt, scratches, discolouration and the biggest splice in the whole of reel 2.
Do you have any recommendations of a good scene to test for rendering quality? From your description the warehouse fight in reel 4 sounds like a potential, though it's got a lot of odd curved scratches.
Program material is recorded on the other side of this disc...
deleted user
Unregistered
Thanks:
Given thank(s) in post(s)
Really ideally you take both encodes, sync them up, for example in AVISynth and then you just go through the entire film randomly looking at frames and pixel peeping, that's the best way to learn and get a "feel" for it. From there you can then further tweak parameters if you want to do that.
Posts: 393
Threads: 23
Joined: 2019 Oct
Thanks: 385
Given 526 thank(s) in 185 post(s)
Country:
So a quick update:
In the end I've decided to go for 2K. While there's slightly more detail in the raw scans there is not enough to warrant the extra time, money and effort for a 4K restoration.
So I've figured out a good 2K workflow now and I've started restoring my sample scene. I'm still learning the new tools but so far I'm very happy with the results. I'm using a combination of user-guided automated cleanup for static portions of a shot, and manual cleanup for the moving elements. So far on average a shot is taking around an hour to two hours to work on, but a couple have taken more. I hope to be able to show some examples soon.
I've now reached the first splice in the sample scene, and I'm trying to work out the best way forward. For this scene the Blu Ray is not only very soft with a lot of analogue video noise, but it's also heavily cropped. As a result it has only a little more detail than the best DVD source I have. The DVD however isn't as heavily cropped. Rather than taking a hit on quality and losing footage to cropping I started thinking about trying to overlay the cropped Blu Ray over the uncropped DVD and using Dr. Dre's colour correction tool to match it to the existing 35mm parts of the shot.
The problem I'm having just now is that I can't perfectly align the two sources. Using Davinci Resolve I can almost get it, but it doesn't allow non-linear transformations, so I can't get it perfect. I don't have Photoshop, and while Gimp allows non-linear transformations it doesn't allow viewing the difference of overlayed images while you scale them, so it's no good either.
Does anyone have any recommendations on how best to perfectly overlay two sources? I assume it would be best to try and colour match them both to the original source before then overlaying them?
Sorry for all the questions, and thanks in advance for any help.
Program material is recorded on the other side of this disc...