SD video becomes dull when upscaling to HD

[Bencuri]

You can change QTGMC() to QTGMC(FPSDivisor=2) to get 25p. But you will lose half the motion because your source is 25i (50 different fields per second), not 25p.

I know, but I don't like the 50p stuff in case of these. Either 50i or 25p. That smoothness of motion of the 50p looks odd for me. I added SelectEven() right after the QTGMC() line meanwhile, the result seems to be okay, but I don't remember if that is the proper way to use that? As I remember a number should be included in the brackets but I am not sure which one to choose, or leave it empty?

[jagabo]

Keep in mind, when you watched the original interlaced video on a CRT TV you saw 50 different images per second. The clip was shot as interlaced video, not film.

QTGMC(FPSDivisor=2) might be faster than QTGMC().SelectEven().

[Bencuri]

Keep in mind, when you watched the original interlaced video on a CRT TV you saw 50 different images per second. The clip was shot as interlaced video, not film.

QTGMC(FPSDivisor=2) might be faster than QTGMC().SelectEven().

About the perception of movement you may be right here. But evaluating the recording properly may be a challenge for me, because I guess the Laptop LCD doesn't display true 50i, Youtube must be converting it to 25p. Most of my recordings I only watch on the Laptop. It is different when watched on TV with my media player, smoother. But still not like 50p in my opinion.

Earlier we had a large CRT TV for a long time, then a few LCDs. When watching my favourite movies from the 70's 80's they looked fine. Recently we purchased a brand new LG Smart TV, I hate watching those old movies on it. They look so odd. They look like being a modern movie, the image is annoyingly smooth. I guess becuase it outputs everything in 50p. I have been watching for solutions how to set it back to 50i to watch certain things like on the previous TVs, but I did not find it.

When I record myself, I can only record in 50i or 25p. Usually I choose 50i, because when record in 25p, there is always flickering in the image during motion. This DVD however is fine with 25p, just like movies on DVD. I have mentioned this oddity earlier, I don't know why consumer camcorders do this. So to avoid that I only use 25p when the scene is pre-set, and there is little motion only. But sometimes even in that case I have problems.

[Bencuri]

In connection with the speeding up of conversion: I found some info that x264.exe was available with OpenCL support, but I did not find where to get that version, or what the parameter is that switches it on?

[Selur]

For the fun of it, I threw BasicVSR++(model=5<-adds some 'funky' textures *gig*) and DPIR at it (script: https://pastebin.com/kVrMgQ52).
The x264 options:

      --opencl                Enable use of OpenCL
      --opencl-clbin <string> Specify path of compiled OpenCL kernel cache
      --opencl-device <integer> Specify OpenCL device ordinal

were more of a 'would this work/help' (iirc those should be available in all normal builds) and at least for SD content are not worth the overhead (usually encoding gets slower)

Cu Selur

[Bencuri]

For the fun of it, I threw BasicVSR++(model=5<-adds some 'funky' textures *gig*) and DPIR at it (script: https://pastebin.com/kVrMgQ52).
The x264 options:

      --opencl                Enable use of OpenCL
      --opencl-clbin <string> Specify path of compiled OpenCL kernel cache
      --opencl-device <integer> Specify OpenCL device ordinal

were more of a 'would this work/help' (iirc those should be available in all normal builds) and at least for SD content are not worth the overhead (usually encoding gets slower)

Cu Selur

Thanks for the code. But i prefer Jagabo's version, on yours the image is a bit textured.

I have tried the opencl parameter, first it says OpenCL is allowed, recognises my GPU but then quits the conversion with fatal error.

I tried the conversion without that, with prefetch(4), it works, but the core is heating up too much, I switched back to 3 instead.

[Selur]

did a quick test:

ffmpeg -y -loglevel fatal -noautorotate -nostdin -threads 8 -i "G:\TestClips&Co\files\test.avi" -map 0:0 -an -sn -color_primaries bt470bg -color_trc smpte170m -colorspace bt470bg -color_range tv  -pix_fmt yuv420p -vsync 0 -f rawvideo - | x264 --preset veryfast --crf 18.00 --profile high --level 5.1 --ref 3 --direct auto --b-adapt 0 --sync-lookahead 48 --qcomp 0.50 --rc-lookahead 40 --qpmax 51 --partitions i4x4,p8x8,b8x8 --no-fast-pskip --subme 5 --aq-mode 0 --vbv-maxrate 300000 --vbv-bufsize 300000 --sar 1:1 --non-deterministic --range tv --colormatrix bt470bg --demuxer raw --input-res 640x352 --input-csp i420 --input-range tv --input-depth 8 --fps 25/1 --opencl --opencl-clbin "J:\tmp\2024-04-26@16_02_31_6910\x264_lookahead_2024-04-26@16_02_31_6910_01.clbin" --output-depth 8 --output "J:\tmp\2024-04-26@16_02_31_6910\2024-04-26@16_02_31_6910_01.264" -

worked fine here,...

The sample was just for fun (playing around with BasicVSR++ a bit to see what the different models might be good for. )
Here's another one (https://pastebin.com/CuuYtT1i).

Cu Selur

[Bencuri]

did a quick test:

ffmpeg -y -loglevel fatal -noautorotate -nostdin -threads 8 -i "G:\TestClips&Co\files\test.avi" -map 0:0 -an -sn -color_primaries bt470bg -color_trc smpte170m -colorspace bt470bg -color_range tv  -pix_fmt yuv420p -vsync 0 -f rawvideo - | x264 --preset veryfast --crf 18.00 --profile high --level 5.1 --ref 3 --direct auto --b-adapt 0 --sync-lookahead 48 --qcomp 0.50 --rc-lookahead 40 --qpmax 51 --partitions i4x4,p8x8,b8x8 --no-fast-pskip --subme 5 --aq-mode 0 --vbv-maxrate 300000 --vbv-bufsize 300000 --sar 1:1 --non-deterministic --range tv --colormatrix bt470bg --demuxer raw --input-res 640x352 --input-csp i420 --input-range tv --input-depth 8 --fps 25/1 --opencl --opencl-clbin "J:\tmp\2024-04-26@16_02_31_6910\x264_lookahead_2024-04-26@16_02_31_6910_01.clbin" --output-depth 8 --output "J:\tmp\2024-04-26@16_02_31_6910\2024-04-26@16_02_31_6910_01.264" -

worked fine here,...

The sample was just for fun (playing around with BasicVSR++ a bit to see what the different models might be good for. )
Here's another one (https://pastebin.com/CuuYtT1i).

Cu Selur

I used x264.exe not ffmpeg.

[Selur]

look at the whole command line, I used ffmpeg for decoding and piped into x264,....

[Bencuri]

look at the whole command line, I used ffmpeg for decoding and piped into x264,....

Indeed, yet for me it does not work, it reports fatal error with OpenCL

But not a big problem, I can investigate it later. The code that jagabo gave works fine. The conversion time is tolerable, so with that I can finish the project with good results. Thanks for the contribution of both of you.

[Bencuri]

Now I am working on a new video from the same DVD series (attached below), but on the recent volume the videos are different. No need to stretch the image as I can determine. But what are the correct values then for the upscaling? I assume it should be targeted to 1350x1080 to avoid stretching. But maybe I am wrong.

I also tried adding a letterbox to the 1350x1080 image with the following code: AddBorders(285, 0, 285, 0). But when I add it there is an error reported: AddBorders:YUV image can only add by Mod 2 (left side)

This was the code previously for this project, this one should be modified to match the image size of the attached video better:

LWLibavVideoSource("Primus.mpg", cache=false, prefer_hw=2) 

ColorMatrix(mode="rec.601->rec.709")
QTGMC()
Dehalo_alpha(rx=2.0, ry=1.0, darkstr=2.0, brightstr=2.0)
SMDegrain(thsad=1000, tr=3, prefilter=4)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1280, fheight=720)
aWarpSharp2(depth=5)
CAS(0.5)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1920, fheight=1080)
aWarpSharp2(depth=5)
CAS(0.5)

prefetch(4)

[jagabo]

You should crop the source to 704x576, then upscale to 1440x1080. Add 240 column black borders at the left and right if you need a 1920x1080 frame size. Otherwise leave it at 1440x1080.

LWLibavVideoSource("Primus.mpg", cache=false, prefer_hw=2)
Crop(12,0,-4,-0)

ColorMatrix(mode="rec.601->rec.709")
QTGMC()
Dehalo_alpha(rx=2.0, ry=1.0, darkstr=2.0, brightstr=2.0)
SMDegrain(thsad=1000, tr=3, prefilter=4)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=960, fheight=720)
aWarpSharp2(depth=5)
CAS(0.5)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1440, fheight=1080)
aWarpSharp2(depth=5)
CAS(0.5)

AddBorders(240,0,240,0)

prefetch(4)

[hello_hello]

It looks like the "Gambala cut2" video is flagged as 4:3, but it's wrong.

If it has a 16:11 pixel aspect ratio, by the time you crop the crud it's virtually 16:9, so I wouldn't bother with borders and just crop it to 16:9. There's a sample encode attached.

Crop(10,0,-6,-0)
QTGMC()
ColorMatrix(mode="rec.601->rec.709", clamp=0)

Dehalo_alpha(rx=2.0, ry=1.0, darkstr=2.0, brightstr=2.0)
SMDegrain(thsad=1000, tr=3, prefilter=4)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1280, fheight=720)
aWarpSharp2(depth=5)
CAS(0.5)

nnedi3_rpow2(2, cshift="Spline36Resize")
CropResize(1920,1080, 12,8,-10,-4, Resizer="Resize8")
aWarpSharp2(depth=5)
CAS(0.5)

Edit: I didn't initially realize jagabo was replying to your question regarding the "primus part" sample in his post above. I missed that you'd asked about a second video. The "primus part" sample is definitely 4:3.

If it was me, I'd still crop a bit more to clean the picture edges and remove the crud top and bottom though.

Crop(10,0,-6,-0)
QTGMC()
ColorMatrix(mode="rec.601->rec.709", clamp=0)

Dehalo_alpha(rx=2.0, ry=1.0, darkstr=2.0, brightstr=2.0)
SMDegrain(thsad=1000, tr=3, prefilter=4)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=960, fheight=720)
aWarpSharp2(depth=5)
CAS(0.5)

nnedi3_rpow2(2, cshift="Spline36Resize")
CropResize(1920,1080, 20,8,-10,-6, CropDAR=4.0/3.0, Resizer="Resize8", Borders=true)
aWarpSharp2(depth=5)
CAS(0.5)

[Attachment 79464 - Click to enlarge]

Personally I prefer de-interlacing to 50p. It can look a bit soap-opera-like when you're not used to it, but I have a fair bit of old video de-interlaced to 50p and after watching it for a while, it bothers me how "jittery" film can look until my brain gets used to it again.
Personally I just crop and resize to square pixel dimensions and upscale on playback, but each to their own....

I've still got an old CRT I don't use any more, but I did fire it up a while ago to use temporarily in the family room. The first thing I noticed was the flicker. When I was watching the CRT regularly, before I bought a HD TV, the flicker wasn't very obvious. I guess my brain got used to it.

[Bencuri]

You should crop the source to 704x576, then upscale to 1440x1080. Add 240 column black borders at the left and right if you need a 1920x1080 frame size. Otherwise leave it at 1440x1080.

LWLibavVideoSource("Primus.mpg", cache=false, prefer_hw=2)
Crop(12,0,-4,-0)

ColorMatrix(mode="rec.601->rec.709")
QTGMC()
Dehalo_alpha(rx=2.0, ry=1.0, darkstr=2.0, brightstr=2.0)
SMDegrain(thsad=1000, tr=3, prefilter=4)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=960, fheight=720)
aWarpSharp2(depth=5)
CAS(0.5)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1440, fheight=1080)
aWarpSharp2(depth=5)
CAS(0.5)

AddBorders(240,0,240,0)

prefetch(4)

Thank you, I tried it, it works very well.

[Bencuri]

It looks like the "Gambala cut2" video is flagged as 4:3, but it's wrong.

If it has a 16:11 pixel aspect ratio, by the time you crop the crud it's virtually 16:9, so I wouldn't bother with borders and just crop it to 16:9. There's a sample encode attached.

Crop(10,0,-6,-0)
QTGMC()
ColorMatrix(mode="rec.601->rec.709", clamp=0)

Dehalo_alpha(rx=2.0, ry=1.0, darkstr=2.0, brightstr=2.0)
SMDegrain(thsad=1000, tr=3, prefilter=4)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1280, fheight=720)
aWarpSharp2(depth=5)
CAS(0.5)

nnedi3_rpow2(2, cshift="Spline36Resize")
CropResize(1920,1080, 12,8,-10,-4, Resizer="Resize8")
aWarpSharp2(depth=5)
CAS(0.5)

Edit: I didn't initially realize jagabo was replying to your question regarding the "primus part" sample in his post above. I missed that you'd asked about a second video. The "primus part" sample is definitely 4:3.

If it was me, I'd still crop a bit more to clean the picture edges and remove the crud top and bottom though.

Crop(10,0,-6,-0)
QTGMC()
ColorMatrix(mode="rec.601->rec.709", clamp=0)

Dehalo_alpha(rx=2.0, ry=1.0, darkstr=2.0, brightstr=2.0)
SMDegrain(thsad=1000, tr=3, prefilter=4)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=960, fheight=720)
aWarpSharp2(depth=5)
CAS(0.5)

nnedi3_rpow2(2, cshift="Spline36Resize")
CropResize(1920,1080, 20,8,-10,-6, CropDAR=4.0/3.0, Resizer="Resize8", Borders=true)
aWarpSharp2(depth=5)
CAS(0.5)

[Attachment 79464 - Click to enlarge]

Personally I prefer de-interlacing to 50p. It can look a bit soap-opera-like when you're not used to it, but I have a fair bit of old video de-interlaced to 50p and after watching it for a while, it bothers me how "jittery" film can look until my brain gets used to it again.
Personally I just crop and resize to square pixel dimensions and upscale on playback, but each to their own....

I've still got an old CRT I don't use any more, but I did fire it up a while ago to use temporarily in the family room. The first thing I noticed was the flicker. When I was watching the CRT regularly, before I bought a HD TV, the flicker wasn't very obvious. I guess my brain got used to it.

I noticed the crud at the top, but indeed, the bottom has an irregularity as well. Thanks for highlighting that. I will cut them out like you suggest.


"soap-opera-like" - haha, indeed. That is exactly how those upgraded old movies look like. But in the end I stuck to 50p in this project as well. There is too much filcker here indeed, so in the end I dropped the 25p idea.

[hello_hello]

QTGMC has a few arguments for adding motion blur you can enable when de-interlacing to 25p. I still prefer 50p though, and the motion blur created by QTGMC isn't quite the same as the real thing, but one reason video de-interlaced to 25p can look a bit jittery, is video doesn't have the same amount of natural motion blur as film, and the motion blur helps film look a bit smoother even though it's only 24fps. Less motion blur is no doubt one of the reason for video looking a bit soap-opera-like, compared to film.

[Bencuri]

QTGMC has a few arguments for adding motion blur you can enable when de-interlacing to 25p. I still prefer 50p though, and the motion blur created by QTGMC isn't quite the same as the real thing, but one reason video de-interlaced to 25p can look a bit jittery, is video doesn't have the same amount of natural motion blur as film, and the motion blur helps film look a bit smoother even though it's only 24fps. Less motion blur is no doubt one of the reason for video looking a bit soap-opera-like, compared to film.

That is good to know, never understood why that flicker is there so many times.

[Bencuri]

I am trying to process the 3rd batch of these videos, but in spite of that they are from they same DVD as the last video I submitted, they seem to be different in dimensions. I am trying to use the last script, but even that way the heads seem to be a bit elongated. Could you help how to modify the sizing here? Or cropping, whichever is needed? I am also not sure in the recent one whether you need the degrain? The original sample does not seem to be too grainy for me.

I have attached the recent sample file and this is the latest code that you suggested.

LWLibavVideoSource("Obligatoire_part.mpg", cache=false, prefer_hw=2)
Crop(12,0,-4,-0)

ColorMatrix(mode="rec.601->rec.709")
QTGMC()
Dehalo_alpha(rx=2.0, ry=1.0, darkstr=2.0, brightstr=2.0)
SMDegrain(thsad=1000, tr=3, prefilter=4)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=960, fheight=720)
aWarpSharp2(depth=5)
CAS(0.5)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1440, fheight=1080)
aWarpSharp2(depth=5)
CAS(0.5)

AddBorders(240,0,240,0)

prefetch(4)

[jagabo]

They definitely look too thin at 4:3. But they look too fat at 16:9. Half way in between looks about right.

[davexnet]

Yes, it's a matter of judgement, this looks closer to me

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1040, fheight=720)  # this one not so important but trty to match ratio
aWarpSharp2(depth=5)
CAS(0.5)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1560, fheight=1080)  # increase the width
aWarpSharp2(depth=5)
CAS(0.5)

AddBorders(180,0,180,0)  # reduce the L/R borders to account for increased width above to still give 1920/1080

[Bencuri]

Yes, it's a matter of judgement, this looks closer to me

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1040, fheight=720)  # this one not so important but trty to match ratio
aWarpSharp2(depth=5)
CAS(0.5)

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1560, fheight=1080)  # increase the width
aWarpSharp2(depth=5)
CAS(0.5)

AddBorders(180,0,180,0)  # reduce the L/R borders to account for increased width above to still give 1920/1080

Thanks! The other problem I noticed with this is that there is some artifact on the right side of the image. It is a few pixels in width. What is the proper method to crop that? I tried using crop plugin before, but it behaved so chaotic that I do not understand at all how it works. Not as I expect. But maybe this is what should be added?:

Crop(0,0,-4,-0)

[Attachment 80124 - Click to enlarge]

The artifact disappears when adding this but I am not sure whether I should also apply cropping the same amount from the left side as well?

[jagabo]

I didn't see that noise in the sample video. But yes, Crop(0,0,-4,-0) removes 4 columns from the right side of the frame.

http://avisynth.nl/index.php/Crop

The first two values are always positive and indicate the amount to remove from the left and top edges of the frame. The second two can be positive or negative. If they are positive they are the width and height of the frame to keep. If they are negative they are the amount to remove from the right and bottom edges of the frame.

There's no need to crop from the left side unless you see junk there too.

By the way, I think you are using way to much noise reduction with SMDegrain(). I'd recommend something around thsad=75, tr=2.

[Bencuri]

The second two can be positive or negative. If they are positive they are the width and height of the frame to keep. If they are negative they are the amount to remove from the right and bottom edges of the frame.

I guess this is the reason why I experienced it works so strange.

I will try with less degrain, thanks!

[hello_hello]

You guys must be on much better drugs than I can get at the moment. I'm sure it's 16:9.

Bencuri,
You might want to try this function while playing with Avisynth's cropping.
https://forum.videohelp.com/threads/395870-Avisynth-and-VapourSynth-cropping-preview-functions

I cropped a bit more from the sides as the amount of crud increases and decreases, especially on the left side.

"Obligatoire_part.mkv"

QTGMC()

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1280, fheight=720)
aWarpSharp2(depth=5)
nnedi3_rpow2(2, cshift="Spline36Resize")
aWarpSharp2(depth=5)

CropResize(1920,1080, 20,0,-10,0, InDAR=16.0/9.0, Resizer="Resize8", ColorMode="601-709")

I tried to resist but Avisynth has a crop function, there's stuff that shouldn't be in the frame and culling most of it reduces the size of the letterbox borders. Everybody wins.

"Obligatoire_again.mkv"

CropResize(1920,1080, 100,0,-50,0, InDAR=16.0/9.0, Resizer="Resize8", ColorMode="601-709")

[Bencuri]

You guys must be on much better drugs than I can get at the moment. I'm sure it's 16:9.

Bencuri,
You might want to try this function while playing with Avisynth's cropping.
https://forum.videohelp.com/threads/395870-Avisynth-and-VapourSynth-cropping-preview-functions

I cropped a bit more from the sides as the amount of crud increases and decreases, especially on the left side.

"Obligatoire_part.mkv"

QTGMC()

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1280, fheight=720)
aWarpSharp2(depth=5)
nnedi3_rpow2(2, cshift="Spline36Resize")
aWarpSharp2(depth=5)

CropResize(1920,1080, 20,0,-10,0, InDAR=16.0/9.0, Resizer="Resize8", ColorMode="601-709")

I tried to resist but Avisynth has a crop function, there's stuff that shouldn't be in the frame and culling most of it reduces the size of the letterbox borders. Everybody wins.

"Obligatoire_again.mkv"

CropResize(1920,1080, 100,0,-50,0, InDAR=16.0/9.0, Resizer="Resize8", ColorMode="601-709")

It is interesting because I did have some feelings that the frame could be wider sometimes, but other times it seemed to be proper. I looks however as if the frame size would have differed originally between certain scenes. I know these guys on the video and sometimes their face has the proper shape, sometimes not. Maybe there was some chaos during the editing. Yo don't have the full video, of course, but I have and the frame size does seem to be chaotic. The version that jagabo shared seems to be a good compromise, but I need to do a full reconverting tonight, so I will test yours as well. Because for fist sight it seems to be fine. Thanks!

[Bencuri]

You guys must be on much better drugs than I can get at the moment. I'm sure it's 16:9.

Bencuri,
You might want to try this function while playing with Avisynth's cropping.
https://forum.videohelp.com/threads/395870-Avisynth-and-VapourSynth-cropping-preview-functions

I cropped a bit more from the sides as the amount of crud increases and decreases, especially on the left side.

"Obligatoire_part.mkv"

QTGMC()

nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1280, fheight=720)
aWarpSharp2(depth=5)
nnedi3_rpow2(2, cshift="Spline36Resize")
aWarpSharp2(depth=5)

CropResize(1920,1080, 20,0,-10,0, InDAR=16.0/9.0, Resizer="Resize8", ColorMode="601-709")

I tried to resist but Avisynth has a crop function, there's stuff that shouldn't be in the frame and culling most of it reduces the size of the letterbox borders. Everybody wins.

"Obligatoire_again.mkv"

CropResize(1920,1080, 100,0,-50,0, InDAR=16.0/9.0, Resizer="Resize8", ColorMode="601-709")

I tried it, and your sizing seems to be correct, except for the parts are mentioned. But that is not your fault, as I said the sizing seems to vary at certain scenes. In case of the sample it is at 0:10. For that scene your sizing results in overstretched faces. However the problem was that I cut a part where this is not very obvious. That scene repeates a few times in the complete video, and at other points it is much more visible, in the example not that much. But the whole clip has varying sizing, and I see now that was that confused me as well. Here is an example:


[Attachment 80227 - Click to enlarge]

However I do not see the difference in your samples. How do the two differ, I don't get it?

[davexnet]

Who are these guys and what is the DVD?

[Bencuri]

Who are these guys and what is the DVD?

Extra Musica - Obligatoire DVD 2004

[hello_hello]

However I do not see the difference in your samples. How do the two differ, I don't get it?

I cropped more from the sides of the second one which means the function cropped a bit more from the top and bottom to compensate, zooming in a little. You can see past the edges of the white backdrop on both sides in the very first frame. I seem to have deleted the original so this is the first frame of first sample.


[Attachment 80228 - Click to enlarge]

Same frame, second sample.


[Attachment 80229 - Click to enlarge]

I doubt the sandbags holding the backdrop in place were included in this shot for artistic reasons (bottom right corner), although you never know....
I think it's the only shot with anything like that remaining in the second sample but it'd probably crop away too much picture in other shots to get rid of it.


[Attachment 80230 - Click to enlarge]

None of that would've been seen back in the says of CRT TVs due to over-scanning.
https://en.wikipedia.org/wiki/Overscan
Even modern TVs probably still over-scan more often than not unless you disable it. Mine's a plasma so it's fairly old but I'm pretty sure it over-scans by 5% unless you tell it not to, aside from viewing free to air TV when you can't disable it. I don't think it's called over-scanning in it's settings, but I can't remember exactly.

I know what you mean about stretched looking faces in the screenshot you posted but I think it's just a perspective thing due to the angle. It's hard to tell but at least at 16:9 it doesn't look like they all have eating disorders.

[davexnet]

In the early 2000's the UK and perhaps some other places experimented with the 14:9 (1.555) aspect ratio
as a way to "ease" the transition of 4:3 > 16:9. Perhaps this is what happened here.