Episodes – 12+4
Video – 1280×720 @ crf 15
Audio – 2.0 FLAC
Subs – Mazui, Static Subs (ED2, ED5, ED8), CoalKirino (ED1, ED3, Extra TLs)
Torrent
In order to keep the option of doing season 2 available, I felt I needed to fix this series. I fixed a few errors in the subs, and re-encoded to 10-bit. I also did the ED a little different this time, concentrating more on the video and less on the credits. I am still not sure if I will do season 2 or not, but this changes it from a definite “no” to a “maybe”.
@Natsu3
How interesting – the comments section of this has made me do more research and finally understand telecine and interlacing more. Now I know why some of my older DVD rips look like a blurry mess….
For my personal rip, I used the latest version of Handbrake (gasp!) and it did a pretty good job doing a IVTC on the body and ED of the video. The only issue is jerky credits in the beginning of the ED when the framerate ramps up to 29.97fps, but I could care less about watching the ED (it’s there for completeness sake).
@Natsu3
There was actually a pattern change iirc in one of the web previews (right at the end before the episode title came on), but yes other than that it was perfect.
My approach is a semi middle ground (and pretty similar to the bobbing/mvtools way). Field match where I need to, NNEDI3 to return *bottom field (essentially nnedi3(field=-2).selectevery(2,1)), and motion mask the remaining the combed areas (credits).
*Bottom field produces moving credits every frame on field matched video at 30fps.
Take a look at these EDs, I would have made them smaller but I deleted my lossless avis already and these are the only raw encodes i have of the straight ed’s.
http://demo.ovh.com/download/91b57b988225dfe4f5731b23ed8805d0/ep01_bd_ed_1080_raw.mkv
http://demo.ovh.com/download/ebc5df138543343e6fe0555a497f555c/ep02_bd_ed_1080_raw.mkv
These ED’s are a stupid compromise no matter how you look at it and there’s no perfect way to do it.
@Natsu3
Do you remember where that combed frame was and which ep? I don’t recall seeing any when I checked in Yatta nor when I watched the final encode.
@ar
The best way to do this credits is obviously 24 fps decimation with credits interpolation (adenosin’s way). Look at first of your ed’s, around frame 1260, background motion isn’t that soft as it supposed to be.
Yes, vfr would looks better on screenshots since you won’t have any interpolation artifacts.
Yes, you won’t notice any interpolation artifacts while watching if you would do it properly (interpolate only credits, not a background), since they are all small and random at every frame.
Yes, you would notice background motion artifacts if you would do it 30 fps.
“Screenshots” are nothing. “Watching” is everything.
@ar
Episode 01; if you stick to the pattern frame 15527 is a combed n-match. If you used PG, it has most likely been changed in a p-match.
—
Aside from the next ep title, there are no previews within the episode’s m2ts though. I assume you spliced them in before ivtc, in which case there’s a 80% the pattern switches of course.
—
I’m not really following. If you’re going to use nnedi3 (deinterlace) then there is really no point in fieldmatching first; you’re trowing the other set of fields (top fields) away anyway. And there couldn’t possibly be combed frames left afterwards. In the end that’s just deinterlacing resulting in 30fps fluid credits with jerky background. Worst case scenario: a pan that moves along with the credits ep01_bd_ed_1080_raw.mkv @00.50 .
————
@tp7
It’s not “24 fps decimation with credits interpolation” because you obviously can’t process background and credits separately. It’s bobbing the whole picture to 60fps and then interpolating 2 frames out of every 5, resulting in a 24fps video with fluid credits & background and some possible small artifacting.
But I agree that this is visually much less disturbing than jerky video.
@tp7
I’ll throw one together with that area interpolated since the rest of the ED nor the 2nd one don’t require it at all.
@Natsu3
With mt_something (like, dunno, mt_lutspa to separate the place where credits are and then mt_whatever to take areas with credits) you can do it mostly separately, in order to save the background [s]and humanity[/s]. Of couce, it won’t help you to save the credits, but you always can use mo-comp deinterlacing filters and so on. Picture would be mostly ideal, but at least 30fps. :Effort: goes here.
And you can, again, _mostly_, interpolate credits separately since we still do have fields without background motion and different position of credits. That’s the main trick, if you are going to interpolate credits between two fields with different background, artifacts will be horrible.
@ar
And one :effort: for you. =)
@Natsu3
I couldn’t find any combed frames in the decimated video, so most likely I fixed it (don’t have the avi’s to check with yatta anymore) or I decimated it (because it was on the 2nd N match (dupe frame).
Regarding what I said above, here’s a proper explanation.
FieldHint(…)
*30/24fps decimate joining here*
nnedi = NNEDI3(field=0,nsize=2,nns=4,qual=2).MAA()
TFM(PP=6,ovr=”tfm_cmatch_ovr.txt”,clip2=nnedi,MI=45)
*ovr file is just c match forcing so TFM doesn’t do something dumb like try and field match when all I am using it for is combed motion masking.*
ovr file contents: “field = top; 0,0 ccccc”
It just uses tfm’s PP and internal combed motion masking to replace remaining combed areas with the interpolated/deinterlaced bottom field from nnedi3 which are basically the credits only leaving the actual background area outside the combed motion mask intact since it’s been field matched prior.
LOL It sound’s pretty silly I know, but in real world applications for the most part it works pretty well (except obv. those shitty pans!).
But yes it’s practically doing the same as bobbing, except I’m throwing out most of the interpolated frame instead of keeping it and keeping most of the field matched frame.
Though in the case of combined horizontal/vertical motion (side scrolling credits) field=0 with nnedi3 + field=top with tfm can cause really bad trailing artifacts with field matching. It’s just taking advantage of the fact that there is only a single motion vector in play between top and bottom field.
Sorry for the tl;dr. I hope it makes more sense now. It’s one of those crazy ideas I came up with a while back that actually worked well in practice (lol).
@tp7
http://demo.ovh.com/download/5e61fd577d466d6db43abdc9f33004e8/ep01_bd_ed_1080v2_raw.mkv
That’s with the middle section 24fps interpolated. Didn’t touch ED2 since iirc it was pretty much perfect. There are artifacts on same frames but as you said, it’s not really noticeable during playback. I probably didn’t need to do the last 2 scenes of interpolation because they weren’t “too” bad but considering the amount of time encoding the lossless avi’s for the raw (3hrs I think it was) I won’t bother with another for now.
@ar
I have no idea what have you done, but you definetly done it wrong. Credits motion is still jerky, and interpolation artifacts are horrible.
http://screenshotcomparison.com/comparison/17571
Both frames are interpolated, first is yours, second is mine (adenosin’s looks mostly the same as mine). It’s simple nnedi interpolation without any masktools stuff and motion interpolation with MFlowInter.
Of course, I won’t give you complete script since it won’t be interesting for me, but as I see now, you are trying to interpolate motion between wrong frames. What you should do is bob every field of your source ( 5 frames (10 fields) -> 10 frames), then interpolate it to double framerate (10 frames -> 20 frames) and then selectevery with some constant pattern like (20,0,5,10,15) (20 frames -> 4 frames). 2 of 4 frames would be interpolated in the end. Be sure to choose the right pattern that would give you less artifacts (as I said before, interpolate credits between frames without any background motion). That’s what I and adenosin have done. You can check THORA’s release and try yourself in reverse engineering.
I’m interesting in what you would finally achieve, so please don’t give up. 😉
I only double-rated it before. I’ve been playing around with it double-double-rated now and the results seem to be cleaner with mflowinter on the 150 odd frame sample I was playing with. Now to wait a few hours to interpolate the whole 740 odd frames and go from there. I’ll report back once I have something worth reporting 😉
@Natsu3
“(Real interlacing and telecining are so different that no encoder in his right mind would ever call a telecined video interlaced.)”
I don’t know if that is a shot at me or not, but I called telecined sources interlaced on blog posts because it is a more common term and more likely to be understood.
@tp7
Take 3~
http://demo.ovh.com/download/5f7b946afe6715dc50c7e063f8803aae/ep01_bd_ed_1080v2_raw.mkv
Looks a hell of a lot better than the last one. Spent a little bit of time reading properly over the mvtools2 docs and playing around. Didn’t spot any major artifacting like before on the motion interpolated frames. Those 2 vertical pans are interpolated to 24fps (2 good/2 int), and the following 2 horizontal are interpolated to 30fps (3 good/2 int). The duplicate frame doesn’t screw with the horizontal pans nearly as much so I just left it in there, the credits play smoother now though much less jitter.
@coalgirls
It’s aimed at anyone, and no one in particular.
Although I do see your intention, just because it’s a more common term doesn’t mean people actually understand it. Especially when using it in a wrong way will only cause more misconceptions like “(telecined) 1080i is only worth 720p” or “(telecined) 1080i and 1080p have different resolutions” and other nonsense like that. Basically you’re only feeding people the illusion of understanding.
Not that I consider this disc worth of 1080p or even 720p btw; with that ridiculous banding, aliasing, interlaced credits and not even a clean ED I wouldn’t even consider it worth taking over the TV releases. I have absolutely no idea why everybody jumps on this show.
@ar
Your replies are mostly confusing because you’re using terms incorrectly.
“But yes it’s practically doing the same as bobbing, except I’m throwing out most of the interpolated frame instead of keeping it and keeping most of the field matched frame.”
You’re not fieldmatching anywhere because 1) you’re using a ccccc override 2) you’re (motion adaptive) deinterlacing. Calling it similar to bobbing (spatial interpolation of both fields) is quite a long stretch too.
“It just uses tfm’s PP and internal combed motion masking to replace remaining combed areas”
You’re not using Tfm’s PP, only its motion mask. And there’s going to be a lot of “remaining” combed areas, since you haven’t done anything yet (remember, you did not fieldmatch with that ccccc override).
You’re better off using a real motion adaptive deinterlacer like TempGaussMC instead of that TFM setup; TFM’s main purpose is not deinterlacing after all.
@tp7
“With mt_something (like, dunno,”
If you can’t even tell how it was done exactly, how can you say for sure the credits/background were processed separately?
By scrolling frame per frame through aden’s encode it’s obvious that the background wasn’t left untouched; every one of two frames is blurred a tiny bit while the other frame retains most of the original aliasing. Check frames 36506+ in ep02, the hairlines under her right hand.
I’ve never seen a telecined BD worth 1080p.
@Natsu3
I didn’t say that adenozin used it, I just said that it’s possible. Though for such an obvious upscale as oreimo is mostly useless. Especially for ED’s, wich has motion in every frame, you won’t notice anything at all. Though it can be usefull for 00006 and 00007.m2ts (another versions of ED’s).
http://screenshotcomparison.com/comparison/17765
First pair is simple nnedi3 interpolation, second – nnedi3 interpolation for credits and tfm for other parts of a picture, done with a simple mt_lutspa mask – mt_lutspa(“absolute”,expr=”y 800 >= y 1000 <= & 255 y 800 y 1050 < & y 1050 – abs 5.1 * 0 ? ? ?") ( http://img190.imageshack.us/img190/9700/58950287.png ). Look on that horizontal line on the chair.
P.S. Motion interpolation artifacts have nothing to do with deinterlacing. 😛
—–
@ar
Yeah, far much better, but pattern you choose is still slightly incorrect.
http://img196.imageshack.us/img196/4629/42509576.png
http://img31.imageshack.us/img31/2580/71657823.png
http://img143.imageshack.us/img143/6136/73662698.png
http://img535.imageshack.us/img535/4966/94089754.png
http://img253.imageshack.us/img253/5679/45184090.png
http://img684.imageshack.us/img684/3972/15617462.png
http://img810.imageshack.us/img810/1795/89687407.png
http://img194.imageshack.us/img194/8629/14316395.png
http://img833.imageshack.us/img833/2088/69752958.png
http://img214.imageshack.us/img214/7198/42952860.png
First group is yours, second is mine. Frames 1459-1463 of your encode. You probably can notice that huge jump between 3rd and 4th frames, which not a problem for my groups since distance between all frames is constant. That's why your credits looks slightly more jerky, than mine or adenozin's. And yes, like Natsu3 already mentioned, your deinterlacing way is wrong. Btw, tgmc could give you not only good credits, but some blending too, so use it on your own risk (that's why I use nnedi here).
Using motion interpolation when the background isn’t a pan, zoom or any other fluid motion is not needed. Just keep the section at 30fps (vfr), crop the credits and deinterlace them, and fieldmatch the remaining picture. Because its only 12fps animation at best, it’s safe to drop decimation. You only have to override every n-match with movement that crosses the credits to a c+ match, in order to match fields with the deinterlaced crop.
—
oclip=last
FieldHint(with necessary overrides)
a=crop(0,0,0,-256)
b=oclip.nnedi3().crop(0,824,0,-80)
c=crop(0,1000,0,0)
stackvertical(a,b,c)
—
I don’t see the use of that mask, it does not take out only the credits but the background they scroll over as well. It’s no different than just cropping.
Crop will give you some little artifacts on some frames (around the line 824 and 1000), that’s why it isn’t a binary mask. Though you can use crop for speed increasing (crop it before call to nnedi3). Any other mask isn’t really needed here, because you still won’t notice any nnedi3 interpolation artifacts in a place with credits motion.
But yes, I agree that motion interpolation isn’t really needed here, I’m just too lazy to check all places with pan, zoom and so on.
@Natsu3
I field match with FieldHint() first, then just use TFM motion mask after to deinterlace the credits only with NNEDI3. That’s why TFM ovr is straight C matches (it’s to make sure TFM doesn’t try and field match already field matched frames is all, a safeguard if you would).
–
All TFM does is create the combed motion mask (PP=6) based on the combed pixel detection (MI=45) in this case. It’s low enough to pick up any interlaced credits without accidentally catching other parts of the video unless it’s not field matched prior. It takes from clip2 (nnedi3’d field matched video) the areas in the combed motion mask only. It’s like using Tdeint+TMM just faster (and adequate for this scenario).
–
I’ve never had a problem doing it this way to be honest. But yeah the only issue is since it’s all field matched prior, the credits jump on the N matched frames. Does this make more sense now? Sorry if I explained it badly.
and the NCOP NCED??
Nice, ill switch from thora’s to cg’s
whoa…
A 720p file actually below a gig even more so, below a half a gig…
I am so shocked!! *Clap*
I’ve been waiting for this anime to fully come out before i download the whole thing but are these the only ep’s out right know or the only 720p ones out right know????
Thanks for the great work/uploads
ep 04 has a small problem
on preview part (.. wake ga na… )
Sure is technical jargin up in hurr.
thanks ^^
Thanks for ep’s 5 and 6 ^^
http://screenshotcomparison.com/comparison.php?id=15724
Not much to see in the comparison if you’ve seen the previous one that was linked. Thanks to __ar for the BDMV screenie. To Aru Obaka-san didn’t believe me when I linked these…
Comparisons aside, Thora’s credits were jerky compared to the encode __ar sent me. The differnce was quite noticeable though I only did comparisons with ED1 so take it with a graint of salt.
Can’t wait til someone starts a thread @ Baka for which to accept. At least there’ll be no ‘flac vs lossy codec’ unless kira gets a mention
Kira will get a mention because they are the popular group for this due to the lolcommentary. But unless they’re hypocrites and choose his version for the specials while they deleted my True Tears with specials, he probably won’t get it.
i hope someone else translates the commentaries…the kira fansub one is so obviously wrong in many places that its painful to watch. i wonder if we could get a translator here like you guys did for some other series extras…but then again these commentaries are pretty long so maybe hard to find someone
Kira’s commentary translations are awful, but they’re the only ones who bothered so we have no real choice. Fansubbers seem to routinely dismiss important extras like that, so beggers can’t be choosers i guess.
TY!! been waiting for this.
@ coalgirl
They should of at least kept the extra’s made from you guys into THORA’s offer but that’s asking too much common sense from the stupid moderators at BAKA BT with their delusion of “the best offers”.
Agreed with h, Fansubbers forget to include extra’s at least Kira-subs are including the commentaries. I’m grateful for any extra’s THAT SHOULD BE INCLUDED.
The commentaries are a huge deal with Oreimo development if you cannot understand that, that’s your problem and I’m sure your not really a fan of the series. At least include them RAW even if there are no translators available…
@ rude
do what I did, DL oreimo from CGi, and commentary from Kira, problem solved.
thanks 😀
Episode 09 is messed up. Please fix it.
Can you be any more specific?
When I play the episode, the sutitles & text and signs all show up on the first few moments all over the screen. And then the rest of the video plays without any subs or anything.
Confirmed.
I got that problem too. (ep 09)
Yup~
Needs a v2 of epi 09.
OK, I see what I did. I applied the timecodes file to the subtitles instead of the video. It’ll be fixed in the 1-12 batch.
mind uploading the script in the meanwhile? to mux it ourselves
@coalgirl
Why wait for the 1-12 batch? Just do it now.
Whine whine whine. Wait a month and be happy it isn’t “wait for the final batch”.
You can get the script at http://www.megaupload.com/?d=TVCT6P9B and the timecodes at http://www.sendspace.com/file/lato6r
Fixed Episode 09 available via DDL.
@suiton
why dont you do it now ?
thanks cgi & ChrisK
Don’t you girls know how to use xdelta?
Anyways, thanks for the release! 😀
Love you girls!
Here’s the patch for you:
http://www.mediafire.com/?id80eexv7f9t3ck
I’ve sent it to mediafire ‘coz i like it, if you guys need other host, let me know!
<3
Read our about page.
Actually, I’m feeling sort of nice today, so here’s an explanation.
“We will not: Create patches. We’d rather waste your bandwidth than our time.”
It takes time to make patches by comparing the two versions, typing out the patch maker, getting the patch out, and still having to upload the fixed .mkv to get the right batch files. Since we don’t care about you, we’ll waste your bandwidth instead of wasting our time.
There’s a GUI version too, ya know?
Drag and Drop…
Anyways, I was not complaining on my first post but… ┐(￣ー￣)┌
Thanks Sync works great <3
Hey,
There’s a same problem with preview in episode 8 as was is ep 9. Again wrong thing with timecodes i guess D:
Peace