Psycho-visually optimized RDO & Tre

Psycho-visually optimized RDO & Trellis
The human eye doesn't just want the image to look similar to the original, it wants the image to have similar complexity. Therefore, we would rather see a somewhat distorted but still detailed block than a non-distorted but completely blurred block. The result is a bias towards a detailed and/or grainy output image, a bit like xvid except that its actual detail rather than ugly blocking (see http://x264dev.multimedia.cx/?p=164 and http://forum.doom9.org/showpost.php?p=1144270&postcount=1 for more info). The purpose of Psy RDO is to keep the complexity of an encoded block similar to the complexity of the original block. This way Psy RDO produces an image that looks much sharper and more detailed in many cases (compared to none Psy RDO). It also helps to preserve film grain greatly! Please note that Psy RDO will inherently hurt metrics, such as PSNR and SSIM. As soon as psycho-visual optimizations are involved, the classical metrics become useless! Also note that Psy RDO will work with RDO modes only: If Partition Decision is set to 6 (or higher), then Psy RDO will be on by default, otherwise it will be disabled. In addition to Psy RDO there also is Psy-Trellis now. This is still considered an “experimental” feature and disable by default, but it seems to help greatly for retaining textures in the video. Note that Psy Trellis is based on Trellis quantization. Consequently it will only be effective with Trellis quantization enabled too (Trellis 1 is sufficient now, but 2 will be more effective).

¦Psy RDO Strength: This setting controls the strength of Psy RDO. Note that the latest Psy RDO patch will automatically scale the strength of Psy RDO, based on the quantizer of the frame! Therefore the “strength” stetting is simply an additional factor, which will be multiplied to the internal scaling factor. The default value for Psy RDO Strength is 1.0, which should be sufficient for “Film” material. Using even higher values may introduce artifacts! Furthermore it is recommended to lower Psy-RDO to a value of 0.4 for “Animation” material. This does not mean that Psy-RDO is generally harmful for “Animation” material, you just needs to lower the strength for such material.
¦Psy Trellis Strength: This setting controls the strength of Psy Trellis. The default value is 0.0 currently, so Psy Trellis will be disabled by default. Anyway, it may be beneficial to use Psy Trellis for encoding “Film” material. But be careful! Tests have shown that a value of 1.0 usually is too strong for Psy Trellis. For most sources a value of 0.15 should be sufficient. Even higher values may introduce artifacts! Also using Psy-Trellis for “Animation” footage is not recommended.
¦Note: At the moment Psy Trellis Strength isn't available in Avidemux. The default strength of 0.0 will be used. Patched builds of libx264 may behave differently though.
Luma Quantization Deadzones
¦Intra Luma Quantization Deadzone
¦[TO-DO] If you know what information to put here, please contact us!
¦Inter Luma Quantization Deadzone
¦[TO-DO] If you know what information to put here, please contact us!
Quantization Matrix
¦Flat Matrix: The quantization is the lossy part of video compression: The coefficients will be divided through the quantization matrix and then rounded off. The “Flat Matrix” is the default quantization matrix of the H.264 specifications - all entries are simply filled with 16's. This matrix is known to give pretty good results for a wide range of videos and bitrates. This means subjective quality as well as PSNR values.
¦JVT Matrix: This is the alternative quantization matrix of the H.264 specifications. Tests have shown that the “JVT Matrix” performs poorly, although it's part of the official specifications. Therefore it's highly recommended to not use this matrix, except for testing and comparison! You will be far better off with the “Flat Matrix” in almost any case.
¦Custom Matrix: This setting allows you to load your own quantization matrices. Creating quantization matrices as a complex task an needs a deep understand of how video compression works in detail. So creating new quantization matrices should be reserved to the H.264 gurus. Nevertheless you can find a list of suitable matrices at this and this location. Please note that your “Deblocking Filter” settings have a huge impact on how good/bad a certain quantization matrix performs! Also most custom matrices are targeting a certain bitrate range (e.g. ultra high or ultra low bitrates) and will perform bad outside this range. Last but not least you should not use any custom matrices, except you know what you are doing. In most cases you will get satisfactory results simply by sticking with the “Flat Matrix” (default).
¦Remarks: Now that x264 contains various psycho-visual optimizations (Adaptive Quantization, Psy-RDO, Psy-Trellis) custom quantization matrices have become obsolete! Most of the things that people tried to achieve with custom matrices, such as detail and grain retention, can now be achieved by Psy optimizations in a more sophisticated way. Furthermore the Psy optimizations are tuned for the default flat matrix. So using “extreme” custom matrices may result in undesired effects when Psy optimizations are involved! Therefore we highly recommend to stick with the “flat” matrix, unless you have a very good reason to use a custom matrix.
Quantizer
Quantizer Control
¦Minimum Quantizer: Specifies the minimum quantizer to be enforced. This means every frame will get at least this amount of data loss. The default value is 10, which makes sure no bitrate is wasted on too low quantizers. This value should be okay, even for high quality videos.
¦Maximum Quantizer: Specifies the maximum quantizer to be allowed. This means none of the Frames will get a higher amount of data loss than this. The default value is 51, which is the maximum quantizer possible. So “Max Qp” is not limited by default. Of course the encoder will only go that high when it's really necessary, so don't worry!
¦Maximum Quantizer Step: Specifies how much the quantizer can change between two consecutive frames. The default value is 4. This makes sure that two consecutive frames won't get too different quantizers. If you allow significant greater QP Steps, this might result in visible quality “jumps” between frames, so don't do that.
¦Average Bitrate Tollerance: This setting effects the “Single Pass” bitrate-based mode only. It controls how precise the encoder will hit the target bitrate (or target file size). The goal of “Bitrate Variance” is to get as close as possible to the quality of an CRF mode encode, while still being somewhere near the target file size. A value of 0.0 would restrict the encoder to exactly hit the desired bitrate. The default value of 1.0 allows an discrepancy of 1%, which is still pretty restrictive but already a lot better than pure CBR. Please note that the discrepancy usually should be within a range of 30%. Furthermore CRF mode still gives much better results than the bitrate-based mode and therefore is the recommended method!
¦Factor between I- and P-Frame Quants: This setting controls how much stronger P-Frames will be compressed compared to I-Frames. A value of 1.0 would assign the same quantizers to P-Frames and I-Frames, while the default value of 1.4 assigns 40% higher quantizers to P-Frames (compared to I-Frames). This equals the “I-Frame Boost” option of Xvid. Compressing P-Frames stronger than I-Frames is recommended, as I-Frames serve as the initial reference of a scene and thus have a huge impact on the quality of the following frames. Therefore you should not change the default value, unless you have a very good reason to do so!
¦Factor between P- and B-Frame Quants: This setting controls how much stronger B-Frames will be compressed compared to P-Frames. A value of 1.0 would assign the same quantizers to B-Frames and P-Frames, while the default value of 1.3 assigns a 30% higher quantizer the B-Frames (compared to P-Frames). Compressing B-Frames stronger than P-Frames is recommended, as B-Frames are not referenced by other frames (except for the B-Pyramid), while P-Frames server as reference for the following frames. Therefore you should not change the default value, unless you have a very good reason to do so!
¦Chroma to Luma Quantizer Offset: This setting controls how much stronger the color information (chroma) will be compressed compared to the brightness information (luma). Sometimes it's makes sens to compress the color information stronger than the brightness information, as data loss in the color information is less visible to the human eye than data loss in the brightness information. The specified offset will be added to the chroma quantizers. It can be configured between -12 and +12. The default value is 0 and usually it's recommended to keep the default value! Note that both, Psy-RDO and Psy-Trellis, will lower the offset by one or by two, if enabled. So you may end up with an offset of -4 by using Psy optimizations.
Quantizer Curve Compression
¦Quantizer Curve Compression (%): This setting is also called “qcomp” or “bitrate variability” (not to be confused with bitrate variance). It controls how much the bitrate can fluctuate over the entire video. Setting this to 0% would enforce a constant bitrate stream, while a value of 100% would result in a constant quantizer stream. The default value is 60%, which gives good results for most videos. So do not change the default, unless you have a very good reason to do so! Note that adaptive quantization (AQ) partially replaces the effect of qcomp and x264 will internally raise qcomp to compensate based on the adaptive quantization strength. Also note that using CRF mode together with a qcomp of 100% is technically equivalent to QP mode, except that CRF mode still is able to use AQ (which QP mode can't do). Hence the more y

Результаты (русский) 1:

Психо визуально оптимизирован RDO и решеткиЧеловеческий глаз не просто хотите, чтобы изображение похожа на оригинал, он хочет на изображение, чтобы иметь аналогичные сложности. Таким образом мы бы скорее увидеть несколько искаженная, но по-прежнему подробные блок чем блок неискаженного но полностью размыты. The result is a bias towards a detailed and/or grainy output image, a bit like xvid except that its actual detail rather than ugly blocking (see http://x264dev.multimedia.cx/?p=164 and http://forum.doom9.org/showpost.php?p=1144270&postcount=1 for more info). Psy RDO предназначен для держать сложности кодированного блока аналогичные сложности исходного блока. Таким образом Psy RDO создает образ, который выглядит гораздо острее и более подробные, во многих случаях (по сравнению с ни Psy RDO). Это также помогает сохранить зерна пленки значительно! Пожалуйста, обратите внимание, что по своей сути Psy RDO повредит метрик, таких, как PSNR и SSIM. Как только психо визуальных оптимизации участвуют, классическая метрики становятся бесполезными! Также обратите внимание, что ПСИ RDO будет работать с только режимы RDO: Если раздел решение устанавливается до 6 (или выше), а затем Psy RDO будет по умолчанию, в противном случае оно будет отключено. В дополнение к Psy RDO также ПСИ-шпалеры сейчас существует. Это до сих пор считается функцией «экспериментальных» и отключить по умолчанию, но она, как представляется, сильно помочь сохранению текстур в видео. Обратите внимание, что Psy шпалеры основана на решетке квантования. Следовательно оно только будет эффективным с шпалеры квантования включен тоже (решетка 1 достаточно сейчас, но 2 будет более эффективным). ¦Psy RDO прочность: этот параметр управляет силой Psy RDO. Обратите внимание, что последний патч Psy RDO автоматически масштабируется сила Psy RDO, основанный на квантователь кадра! Поэтому stetting «силы» является лишь дополнительным фактором, который будет умножаться на внутренний коэффициент масштабирования. Значение по умолчанию для ПСИ RDO сила — 1.0, которая должна быть достаточной для материала «Фильм». Использование даже более высокие значения могут ввести артефакты! Кроме того рекомендуется снизить ПСИ-RDO значение 0,4 для материала «Анимация». Это не означает, что ПСИ-RDO вообще вредно для материала «Анимация», вам просто необходимо снизить прочность для такого материала.¦Psy прочность решетки: этот параметр управляет силой Psy шпалеры. Значение по умолчанию равно 0.0 в настоящее время, так что Psy решетка будет отключена по умолчанию. Во всяком случае это может быть выгодно использовать Psy шпалеры для кодирования материал «Фильм». Но будьте осторожны! Испытания показали, что значение 1.0 обычно слишком сильна для Psy шпалеры. Для большинства источников значение 0,15 должно быть достаточно. Даже более высокие значения могут ввести артефакты! Не рекомендуется также использовать ПСИ-шпалеры для «Анимация» кадры. ¦Note: на данный момент ПСИ шпалеры сила не доступны в Avidemux. Будет использоваться по умолчанию сила 0.0. Залатанное строит libx264 может вести себя по-разному хотя.Лума квантования Deadzones¦Intra Лума квантования мертвой¦ [Задачи] Если вы знаете какую информацию поставить здесь, пожалуйста, свяжитесь с нами!¦Inter Лума квантования мертвой¦ [Задачи] Если вы знаете какую информацию поставить здесь, пожалуйста, свяжитесь с нами!Квантование матрица¦Flat матрица: квантование является потерями частью сжатия видео: коэффициенты будут разделены через матрицу квантования и затем округлены. «Плоская матрица» является по умолчанию квантования матрица спецификации H.264 - все просто заполняются с 16 в. Известно, что эта матрица дают очень хорошие результаты на широкий спектр видео и битрейтов. Это означает, что субъективное качество, а также значения PSNR.¦JVT матрица: это альтернативный квантования матрица спецификации H.264. Испытания показали, что «СГП матрица» выполняет плохо, хотя он является частью официальной спецификации. Поэтому настоятельно рекомендуется не использовать эту матрицу, за исключением тестирования и сравнения! Вам будет гораздо лучше с «Плоская матрица» в почти любом случае.¦Custom Matrix: This setting allows you to load your own quantization matrices. Creating quantization matrices as a complex task an needs a deep understand of how video compression works in detail. So creating new quantization matrices should be reserved to the H.264 gurus. Nevertheless you can find a list of suitable matrices at this and this location. Please note that your “Deblocking Filter” settings have a huge impact on how good/bad a certain quantization matrix performs! Also most custom matrices are targeting a certain bitrate range (e.g. ultra high or ultra low bitrates) and will perform bad outside this range. Last but not least you should not use any custom matrices, except you know what you are doing. In most cases you will get satisfactory results simply by sticking with the “Flat Matrix” (default).¦Remarks: Now that x264 contains various psycho-visual optimizations (Adaptive Quantization, Psy-RDO, Psy-Trellis) custom quantization matrices have become obsolete! Most of the things that people tried to achieve with custom matrices, such as detail and grain retention, can now be achieved by Psy optimizations in a more sophisticated way. Furthermore the Psy optimizations are tuned for the default flat matrix. So using “extreme” custom matrices may result in undesired effects when Psy optimizations are involved! Therefore we highly recommend to stick with the “flat” matrix, unless you have a very good reason to use a custom matrix.QuantizerQuantizer Control¦Minimum Quantizer: Specifies the minimum quantizer to be enforced. This means every frame will get at least this amount of data loss. The default value is 10, which makes sure no bitrate is wasted on too low quantizers. This value should be okay, even for high quality videos.¦Maximum Quantizer: Specifies the maximum quantizer to be allowed. This means none of the Frames will get a higher amount of data loss than this. The default value is 51, which is the maximum quantizer possible. So “Max Qp” is not limited by default. Of course the encoder will only go that high when it's really necessary, so don't worry!¦Maximum Quantizer Step: Specifies how much the quantizer can change between two consecutive frames. The default value is 4. This makes sure that two consecutive frames won't get too different quantizers. If you allow significant greater QP Steps, this might result in visible quality “jumps” between frames, so don't do that.¦Average Bitrate Tollerance: This setting effects the “Single Pass” bitrate-based mode only. It controls how precise the encoder will hit the target bitrate (or target file size). The goal of “Bitrate Variance” is to get as close as possible to the quality of an CRF mode encode, while still being somewhere near the target file size. A value of 0.0 would restrict the encoder to exactly hit the desired bitrate. The default value of 1.0 allows an discrepancy of 1%, which is still pretty restrictive but already a lot better than pure CBR. Please note that the discrepancy usually should be within a range of 30%. Furthermore CRF mode still gives much better results than the bitrate-based mode and therefore is the recommended method!¦Factor between I- and P-Frame Quants: This setting controls how much stronger P-Frames will be compressed compared to I-Frames. A value of 1.0 would assign the same quantizers to P-Frames and I-Frames, while the default value of 1.4 assigns 40% higher quantizers to P-Frames (compared to I-Frames). This equals the “I-Frame Boost” option of Xvid. Compressing P-Frames stronger than I-Frames is recommended, as I-Frames serve as the initial reference of a scene and thus have a huge impact on the quality of the following frames. Therefore you should not change the default value, unless you have a very good reason to do so!¦Factor between P- and B-Frame Quants: This setting controls how much stronger B-Frames will be compressed compared to P-Frames. A value of 1.0 would assign the same quantizers to B-Frames and P-Frames, while the default value of 1.3 assigns a 30% higher quantizer the B-Frames (compared to P-Frames). Compressing B-Frames stronger than P-Frames is recommended, as B-Frames are not referenced by other frames (except for the B-Pyramid), while P-Frames server as reference for the following frames. Therefore you should not change the default value, unless you have a very good reason to do so!¦Chroma to Luma Quantizer Offset: This setting controls how much stronger the color information (chroma) will be compressed compared to the brightness information (luma). Sometimes it's makes sens to compress the color information stronger than the brightness information, as data loss in the color information is less visible to the human eye than data loss in the brightness information. The specified offset will be added to the chroma quantizers. It can be configured between -12 and +12. The default value is 0 and usually it's recommended to keep the default value! Note that both, Psy-RDO and Psy-Trellis, will lower the offset by one or by two, if enabled. So you may end up with an offset of -4 by using Psy optimizations.Quantizer Curve Compression¦Quantizer Curve Compression (%): This setting is also called “qcomp” or “bitrate variability” (not to be confused with bitrate variance). It controls how much the bitrate can fluctuate over the entire video. Setting this to 0% would enforce a constant bitrate stream, while a value of 100% would result in a constant quantizer stream. The default value is 60%, which gives good results for most videos. So do not change the default, unless you have a very good reason to do so! Note that adaptive quantization (AQ) partially replaces the effect of qcomp and x264 will internally raise qcomp to compensate based on the adaptive quantization strength. Also note that using CRF mode together with a qcomp of 100% is technically equivalent to QP mode, except that CRF mode still is able to use AQ (which QP mode can't do). Hence the more y