Help me understand MTS/AVCHD

[Mylo]

I just got my new camera, GH2, and I love it. What I don't get about it is the coded-file format. I'll be honest, I really don't know much about codes in the first place.

So far what I understand is this - the camera outputs a MTS file wrapped in the AVCHD container. It works like a charm in programs like Premiere Pro.

But if you want it to work in some more serious video editing/color correction programs, you need to convert the files to formats like Apple ProRes.

There's a program I just downloaded called 5D 2 RGB and it converts MTS to Apple ProRes format, and I need it for the program I'm using.

I just finished with my first 5D 2 RGB conversion (settings: APR 422, 709, Full Range)...it converted my 2 GB MTS file (17 minutes RT, 720p 60fps) into a 20 GB MOV.

I can't open the MOV file in Media Player Classic HC, but I can open it in Quick Time. It looks somewhat worse than the original, first thing I've noticed is the red borders in the footage are bleeding. In the original file, opened in MPC-HC (with "video mixing rendered 9", if that makes any difference) the red borders look just fine.

On the other hand, it appears to me that the shadows in the ProRes file are not as crushed as in the original MTS.

What did the 5D to RGB do to the MTS, it's 10x the size but the quality is down a notch...am I supposed to lose info within my pixels like this, or is there a safer workflow when moving from one file type to another?

[poisondeathray]

So far what I understand is this - the camera outputs a MTS file wrapped in the AVCHD container. It works like a charm in programs like Premiere Pro.

MTS is the container. It's a "transport stream"

But if you want it to work in some more serious video editing/color correction programs, you need to convert the files to formats like Apple ProRes.

Not necessarily true. It's more dependent on what platform and software you are using. If you're using CS5/CS6 it converts everything to 32bit float internally, higher precision than 10bit prores. This is probably misinformation from old FCP mac users who basically needed to convert to ProRes and couldn't edit natively. Prores is a lossy format (you lose quality, albeit very tiny amount if you use HQ or higher)

One difference is the type of chroma upsampling . In the past, upsampling from YV12 (this is 4:2:0 AVCHD format uses), to RGB in the editor causes "blocky" color edges, because they used a very simple "nearest neighbor" algorithm. This is not the case in newer editors, the sampling is much improved. Still some people prefer 5D2RGB for chroma upsampling. Instead of "blocky" color edges, essentially it' s more smoothed or blurred . There is no "perfect" way of chroma upsampling, all methods have pros/cons

The other difference is the ease of editing. ProRes is less compressed I-frame only format. Performance is going to be a lot smoother and faster (especially on a Mac ; on a PC the differences aren't as big, because PC's don't handle prores as nicely as FCP's native format) . AVCHD is highly compressed and uses long GOP. Seeking is much slower

I can't open the MOV file in Media Player Classic HC, but I can open it in Quick Time. It looks somewhat worse than the original, first thing I've noticed is the red borders in the footage are bleeding. In the original file, opened in MPC-HC (with "video mixing rendered 9", if that makes any difference) the red borders look just fine.

It might be from the 5D2RGB chroma upsampling or your player settings, or combination of things . You're not doing an apples to apples to comparison, so it's difficult to say.

On the other hand, it appears to me that the shadows in the ProRes file are not as crushed as in the original MTS.

This is because of the a) full range conversion b) quicktime uses different gamma than everything else. (~1.8 , the rest of the world uses ~2.2)

The black level isn't crushed, it's just streched out in the full range conversion. Instead of "mapping" Y' 16-235 to RGB 0-255 , Y' 0-255 gets "mapped" to 0-255 . So everything looks washed
out with less contrast.

Native AVCHD is handled as YUV in Premiere . This means that information that you think you recovered from the full range conversion is still accessible (nothing is clipped below Y' 16 or above Y'235). This means if you load the original files, you can color correct and access all the information by using the YUV labelled filters . However if you convert to some RGB format like a tiff sequence in standard range , that data will be clipped

What did the 5D to RGB do to the MTS, it's 10x the size but the quality is down a notch...am I supposed to lose info within my pixels like this, or is there a safer workflow when moving from one file type to another?

To reiterate - People primarily use this for the either the full range conversion (there are other ways to do this), conversion to prores because they need or want it (e.g. ease of editing), or for different chroma sampling (less blocky color edges)

[Mylo]

I'm really really thankful for your indepth post...thank you.

One question if i may...the program I need to use doesn't support MTS so the conversion is inevitable. If I don't go down the ProRes conversion route, I can go and convert my MTS to H.264

But that basically means I'm losing data, no matter how high quality setting or high bitrates per seconds I export?

[poisondeathray]

I'm really really thankful for your indepth post...thank you.

One question if i may...the program I need to use doesn't support MTS so the conversion is inevitable. If I don't go down the ProRes conversion route, I can go and convert my MTS to H.264

But that basically means I'm losing data, no matter how high quality setting or high bitrates per seconds I export?

Which program is that ?

Technically, yes all "flavours" of ProRes are lossy, including ProRes4444 so yes, you do lose some information and there are some compression artifacts generated. But they are "visually lossless." Meaning, very tiny , barely noticable compression losses. You won't be able to tell the difference unless you go frame by frame and maybe zoom in a bit . It can undergo several generations before you notice any deterioration (unless you pixel peep)

But it sounds to me that you're more concerned about the visual differences in levels and chroma sampling due to your workflow or method of viewing - that' s not "prores's" fault - that's a workflow or display configuration issue. Prores encoder (5D2RGB uses ffmpeg libraries behind the scene) can only encode what is given to it. If you've did some manipulation before it even gets encoded, that's not "prores" codec's fault

[Mylo]

The program is Davinci Resolve, it's an amazing piece of software but it doesn't handle MTS in any way so I need to resort to conversion.

So far I'm aware of ProRes conversion and H.264 conversion...both work, what I would like is to go the way that doesn't lose too much video info.

[poisondeathray]

Most people using resolve use clipwrap to rewrap into MOV (mac users at least). This is identical video, just different "package" . It's like putting the audio, video & accessory data into a different "box" . No quality loss

If you 're on windows, you can batch rewrap using ffmpeg or ffmbc . If you don't know how to use the command line, here is a GUI for this process

http://personal-view.com/talks/discussion/4775/update-0.3-beta-winrewrap-batch-rewrap-...pport.#Item_33

EDIT: It looks like he might be charging for it. If he is, I can help you with the command line (ffmpeg and ffmbc are free) . But it might be worth it to you for the GUI and ease of use



As a new GH2 user, you should really read up on the GH2 firmware "hacks". Dramatically improved quality , bitrates and other features

[Mylo]

I'm aware of the hacks, didn't decide to go down those routes, mainly because I don't want to pixel peep...I simply need to learn about making films first, then I'll stress about quality.

I noticed one thing while editing - large pixel blocks that appear if I deliberately blow up the gamma or play with RGB curves...will those blocks go away if I up the bitrate?

I've made a decision with myself - once I start making real $ with my camera, I will probably get me an GH3 and hack my current GH2...until then, I'll deal with this. As I always say - my favorite movie of all time (Apocalypse now) is grainy with loads of chromatic aberrations and barrel distorsions - and it still is the best in my eyes

[poisondeathray]

I noticed one thing while editing - large pixel blocks that appear if I deliberately blow up the gamma or play with RGB curves...will those blocks go away if I up the bitrate?

If you're referring to the source - most of the time these are from compression artifacts . You usually see them more prevalent in dark , low luminance areas. Yes, most of them go away with higher bitrates (ahmm check out some of the hacks) . But that's still a "band aid" approach to adequate and proper lighting . Good lighting tends to fix many of the problems even with unhacked footage

I've made a decision with myself - once I start making real $ with my camera, I will probably get me an GH3 and hack my current GH2...until then, I'll deal with this. As I always say - my favorite movie of all time (Apocalypse now) is grainy with loads of chromatic aberrations and barrel distorsions - and it still is the best in my eyes

Yes, content & story are still "king" in my eyes . The other "technical" things take a backseat

Good luck with the film making

[poisondeathray]

Just to add - you can only grade and "push" 8bit compressed AVCHD footage so far. It tends to "fall apart" quite easily. The hacked footage you can definitely push farther, because fine noise and grain is retained (especially in areas like shadows, dark areas) . It acts almost like dithering effect, whereas the unhacked footage tends to show those ugly compression artifacts a lot sooner. But even that is pale in comparison to higher quality ,higher bitdepth, higher bitrate footage . Or even "raw" footage like from the BlackMagic cinema camera, or Red . You don't have as much "wiggle" room with GH2 footage, so your shot has to be planned more closely to the desired result . But with other higher bitdepth footage, you have a lot more room to make adjustments and changes in the grading suite in post

[Mylo]

I'm aware of that, and to be honest, even though Coppola voted for the GH2 in the Zacuto shootout and I got geeked about it (Coppola voting for "my" camera lol) I still have a hard time noticing the difference between stock and hacked GH2 when I'm looking "on the fly".

Maybe it's because I'm not coming from a video background so the moment I start watching someone's film, I start looking at the things I care about out of my own curiosity - how did they light the scene, how did they make the transitions between shots and dialogues, how did they make the camera move this way or that way, are the talents believable or are they corny, how did they design the sound effects...

Pixel peeping is something I never ever bother with...I think the only thing that "bothers" me DSLR video-wise is back few years ago when 5DMK2 was the go-to camera, I never really understood why 95% of people who make videos with that camera always make (or don't fix) the yellowish tone of the footage...but I never got into shadows vs highlights comparisons

[poisondeathray]

I'm aware of that, and to be honest, even though Coppola voted for the GH2 in the Zacuto shootout and I got geeked about it (Coppola voting for "my" camera lol) I still have a hard time noticing the difference between stock and hacked GH2 when I'm looking "on the fly".

Yes that was too funny

On the fly, yes the comparisons can be hard to see the differences . The biggest differences are in the shadow detail , and if you have lots of motion, or it's a difficult to compress content - the unhacked footage will fall apart more

I think if you start grading more footage in resolve, you will inadvertenly become a "pixel peeper" . Welcome to the club!

Does it make you a better film maker? or story teller ? - well not really - but you really see these differences in resolve (or any grading application) .

But for the price range, the overall image quality really is very good even with the stock firmware. There are price drops all over now that GH3 is out

Maybe it's because I'm not coming from a video background so the moment I start watching someone's film, I start looking at the things I care about out of my own curiosity - how did they light the scene, how did they make the transitions between shots and dialogues, how did they make the camera move this way or that way, are the talents believable or are they corny, how did they design the sound effects...

Haha me too . I almost have to watch it a few times : once for "fun" and a few other times because I was curious on how the did that camera move or something like that

Pixel peeping is something I never ever bother with...I think the only thing that "bothers" me DSLR video-wise is back few years ago when 5DMK2 was the go-to camera, I never really understood why 95% of people who make videos with that camera always make (or don't fix) the yellowish tone of the footage...but I never got into shadows vs highlights comparisons

I hated that camera for video . The aliasing made it unusable IMO. But great stills camera.

[budwzr]

I'm aware of that, and to be honest, even though Coppola voted for the GH2 in the Zacuto shootout ...

Walter Murch rated FCP #1 too, then Apple dumped it.

[Mylo]

I know I know the Coppola thing doesn't mean much, but it makes me warm around the heart and that's what matters I often fantasize about Apocalypse Now: Again! - the 3D Sequel shot entirely on GH2 with 3D 12.5mm, co-produced by Mike Bay, would've had all kinds of explosions and the footage would look like something out of 1993 video game intro

All those helicopter motion + fire artifacts, lil GH2 wouldn't exactly have a field day with that

But on a serious note, I think the GH2 weaknesses are in fact - its strenghts. Just like Air Jordan used to say GH2 will have a hard time dealing with certain situations so its on you to learn how to overcome.

Proper lightning will always help, so I NEED to learn lightning, because I obviously can't rely on camera's dynamic range (not so great) or ISO...so from that perspective, there's something good to its weaknesses. Overall, for the price I paid I really can't complain, sometimes I feel the camera was underpriced, I got the 14-140 kit for 800...I'm not allowed to complain

Also, I think if anyone is serious about movies, you can't rely solely on GH2...especially nowadays when they're selling Black Magic.


One question though - with all those hack going around, why can't someone figure out how to get the raw footage out of the camera, or at least how to get the non-AVCHD compressed stuff on the SD card? Is such thing even possible or did Panasonic made it impossible to work around?

[poisondeathray]

One question though - with all those hack going around, why can't someone figure out how to get the raw footage out of the camera, or at least how to get the non-AVCHD compressed stuff on the SD card? Is such thing even possible or did Panasonic made it impossible to work around?

"Raw" is impossible to get from this camera. Raw is undebayered sensor data - there is no way to access this in the GH2

The short version - actually someone has managed to get uncompressed out of the HDMI with an external reorder - Ralph B - before VK came out with the hack. If you recall all the other canon DSLRs at the time didn't have clean HDMI signals - they were either cropped or riddled with overlays - unusable. But there are problems . IMO the hack superceeds this method because of ease of use

The HDMI out isn't "clean" either in signal or cadence . It should be 4:2:2 uncompressed 59.94 Hz . Theoretically , even at 8bit it would be better than the hack - and it is - in the sense that there is no compression. However, it seems Panny deliberately screwed and output a random 60.0Hz signal , there are fluctuations with inserted fields . Also the chroma is messed up - Once you "fix" it with filters the equivalent chroma resolution is almost no different than 4:2:0 - it's been measured by chroma resolution charts . The "fixes" were done with avisynth scripts with adaptive field matching and downsampling the chroma . It's not perfect, because motion interpolation is used to fill in a missing frame every so often - sometimes that interpolated frame was perfect, sometimes it was noticably "off"

These early days were actually where you say the differences between AVCHD recording and uncompressed. The "hack" is sort of in between these in terms of quality , but has none of the issues with the HDMI workarounds. It sounds ironic that a "hack" is more reliable.. anyway...

[poisondeathray]

I know I know the Coppola thing doesn't mean much, but it makes me warm around the heart and that's what matters I often fantasize about Apocalypse Now: Again! - the 3D Sequel shot entirely on GH2 with 3D 12.5mm, co-produced by Mike Bay, would've had all kinds of explosions and the footage would look like something out of 1993 video game intro

Hahah . Michael fricken Bay + Coppola production. I can see a new film "Transformers Apocalyse Tomorrow"!

It was hilarious to me , because the various GH2 forums were "buzzing" like crazy - and I sort of know how you felt, because I was shooting stuff with the GH2 at the same time .

[Mylo]

Poison you really helped me out with your answers, thank you very much...I'm still googling some of the terms you used, I'm learning.

One thing I've saw on another forums about hacked GH2 was this:

in-camera encoding is mediocre, it's 25p laid on a 50i stream, unflagged, meaning consumer junk.

What does "25p laid on 50i stream" mean? Isn't 25p just 25 progressive frames...or what?

Also, what does "unflagged" mean?

[poisondeathray]

Poison you really helped me out with your answers, thank you very much...I'm still googling some of the terms you used, I'm learning.

One thing I've saw on another forums about hacked GH2 was this:

in-camera encoding is mediocre, it's 25p laid on a 50i stream, unflagged, meaning consumer junk.

What does "25p laid on 50i stream" mean? Isn't 25p just 25 progressive frames...or what?

Also, what does "unflagged" mean?

Not to get too technical (unless you want more info) , the short version -

25p can be encoded either as A) native progressive (known as 25pN) , or B) "25p in a 50i" stream (it's progressive content but encoded as fields in as interlaced) (also know as 25PsF - progressive segmented frames). Many consumer camcorders record the 2nd variety but most pro/sumer camcorders record pN

The ramifications - many softwares, NLE's will get "confused" and deinterlace the 2nd type (thereby degrading the image, reducing the effective resolution practically in half ) because in the video bitstream the field_pic_flag is set to 1 . Unless you take steps to manually interpret the footage, you can seriously degrade your footage . This isn't unique to the GH2 - it's a univeral issue

25pN is "always" better because there is no confusion . "Unflagged" actually means that flag is set to 0 (which is equivalent to progressive) so there is some misinformation in your quote from the other forum . Native progressive streams are handled fine in all software

In both cases the content is the same, its' just that some software handling of the streams get confused and may mishandle the "25p in 50i" case

You're in a "PAL" 50Hz region, but this same thing applies to "30p" (29.97p) streams in 60Hz regions. The 29.97p can be encoded and flagged as interlaced in a 59.94i stream or as native progressive "29.97pN"

[Mylo]

Hey man, thank you x 333

Just to ask - what o you mean when you say "flagged"?

[poisondeathray]

Just to ask - what o you mean when you say "flagged"?

This refers to bitstream flags in the AVC stream . They can convey different types of information to the hardware or software receiving the stream

In the "25p in 50i" case ,

field_pic_flag = 1

It's a binary off/on flag, so 0 essentially means progressive, 1 means interlaced. It's almost like a label or "nametag". It's telling equipment or software: "hey I'm an interlaced (or more precisely, was encoded interlaced) stream". In this specific example, the content of the stream can be different that the method of encoding (e.g. you can have 25p content, but it was encoded as 50i . A true interlaced stream would have 50i content in a 50i stream - that is each field would have a picture representing a different moment in time)

There are other types of flags . For example VUI flags e.g. A full range flag would tell the decoder to decode the YUV data at full range instead of limited range. A Rec601 matrix coefficients flags would tell the decoder it should use ITU Rec.601 matrix for the conversion . Whether or not the receiving hardware or software "obeys" or "disregards" the flag is something else entirely . Not all flags need to take binary on/off format (some can have different values)