Prevents some scale factors from simply not rendering anything at all, resulting in weird scene layouts. While this may incur a higher performance penalty, it does fix the issue with enough accuracy to be deployed into production.
As none of the known Upscaling algorithms handle Alpha, we need to manually restore and interpolate it properly. While this technically reduces visual quality slightly, the chances of this being noticed are slim to none considering that Upscaling is already a questionable solution to quality. Linear was picked here as it produces the best overall result for scaling, keeping gradients mostly in one piece. Mostly.
Fixes#646
As the Alpha channel is completely ignored and possibly destroyed by denoising algorithms, we should restore the Alpha channel manually. Linear interpolation was chosen here as it will behave like Point if the size matches, and properly interpolate if the size doesn't match.
Fixes: #646
The previous name was too strict on what could be put into the effect, and would result in additional clutter in the Filter menu when we eventually decide to support other Upscaling methods than Super-Resolution networks.
As not all encoders need all functions, classes inheriting from this should not need to implement all of them as no-ops. Instead the header should take care of this, which reduces duplicated empty code paths.
As vargs may be modified by some functions, we should not reuse it and instead create a copy of it. This fixes a segfault on logging calls happening with GCC, and potentially may fix other compilers and platforms as well.
While the previous method worked, it matches no other implementation including a reference implementation. The new implementation almost perfectly matches the reference implementation and uses oversampling to achieve the goal. This has the downside of limiting the blur size to just 64, but it is necessary in order to achieve correct results.
Fixes#573
The CUDA library is always available as a singleton, so it does not make sense for it to be passed in. Instead we can simply grab it from the singleton and use it as it is, which makes the code easier to maintain and automates certain code.
This reduces the total amount of links Supporters can submit to just one instead of two. Additionally by removing the buttons and making the entry itself clickable, the UI can show more entries at once and allow users to easily make the connection between the entry and the link it opens when clicked.
The 'obs-ffmpeg-mux.c' file specifies different color parameters than StreamFX does. This causes re-muxing to go haywire, and editors that trust these tags suddenly spew out bad colors for BT.601 and sRGB.
Reverts #478
Qt defaults to give every QAction a TextHeuristicRole, which means that certain key words will cause Qt to change how the QAction behaves. We do not want this, so we explicitly assign it to have NoRole instead.
Fixes#323
Instead of adding ourselves as the last entry which seems to confuse the MacOS Qt implementation, we instead insert ourselves before the Help menu. This should hopefully prevent the StreamFX menu from overriding the OBS About entry.
Fixes#323
Grouping properties by what they do improves the user experience as the user does not have to guess at what belongs to which part of the effect. Additionally toggleable groups automatically disable all child elements, so the user is not confused by them still being active.
While the long descriptions were useful, keeping the updated and translated is pretty much impossible. Technology moves fast and not everyone that translates the project knows a lot about technology.
Therefore the long descriptions have now been replaced with a button that opens the wiki page for the feature instead. This should drastically reduce the number of help cases, and improve the translation coverage at the same time.
Both Rec. 601 and sRGB looked extremely wrong before, resulting in weird or warped colors. Since it is very difficult to find up to date and accurate information on standards, we should simply go for what has the most widespread support.
There is hardly any reason for us to recalculate everything all the time. LUTs can cache the work once, and then re-use it every time necessary, drastically reducing the impact of Color Grading by almost 60% (on some GPUs even more). Additionally this fixes the negative gamma issue, which plagued the filter for a while.
In the future, once PR 4199 (https://github.com/obsproject/obs-studio/pull/4199) has been merged, we can cut away one intermediate rendering step currently required to make the effect work. Hopefully this will be with the 27.x release of OBS Studio.
For simple image and video editing, LUTs (Look-Up Tables) are vastly superior to running the entire editing operation on each pixel - especially if all the processing can be done inside a single shader.
Due to the post-processing requirements for our LUTs, we are limited to 8 bits per channel - though clever use of the unused Alpha channel may result in additional space. For our purposes however, this is definitely enough.
A complete redesign of the component and dependency system is necessary in order to support additional platforms, such as MacOS and other Linux platforms. Additionally it results in a much cleaner code base, which is less confusing overall.
Eventually it might be necessary to push components of StreamFX into their own CMake projects, as it is getting kind of complex now. Especially with the push for a proper plugin manager, things get dicey for big plugins like StreamFX.
Ignoring the data parameter during initialization results in duplication and some third party plugins not working as expected, so it's better to not ignore it.
Fixes#315
Adds support for the AMD Advanced Media Framework H.264 and H.265 encoders via FFmpeg. The majority of settings are supported, and the UI/UX experience mimics that of the NVENC implementation. Various settings are left out due to their complexity and should be controlled via the custom parameters field.
Using the obs_module_file and obs_module_config_path macros works okay, but it comes with a slight overhead as well as additional requirements when passing it to C++ functions that expect certain rules to be fulfilled. By instead wrapping the actual functionality into our own functions and using those we can avoid most of the issues that come with the old approach.
Related: #359
This fixes a bug with older Qt versions which would not remove the modal window from the parent in the setModal() call, resulting in an unusable parent window.
Fixes#368
Improves the previous logic and makes it compatible with the new additions in 26.x, such as sRGB. This was previously broken as the focus was on existing features which could be tested without requiring a compiler to be installed.
Incorrect understanding of how sRGB works with RGB and YCC/YUV formats also caused sRGB to be treated as RGB when I444 was selected. This should also now be fixed, hopefully permanently.
Fixes#331
In order to work around a bug in the OBS Studio UI code, we have to swap our modal status right before showing and right after hiding, so that the OBS Studio tray menu continuous working correctly. This is a bit of a weird solution, but it does work as expected.
Related: https://github.com/obsproject/obs-studio/issues/3518
Implements a manual and automatic update checker with support for both release and testing update channels, allowing users to stay as up to date as possible. It is fully compliant with privacy regulations around the world, as it stays completely silent and inactive until the user gives the Ok to connect to GitHub for the latest releases.
Removes the NVENC entry on systems without an NVIDIA GPU by checking if the library for it can be loaded. If it can't be loaded, it's likely that the user does not have a system with NVENC capabilities - and guaranteed that they can't use the encoder as FFmpeg relies on these libraries.
Adds a utility class and functions to load libraries and symbols from libraries somewhat safely. Libraries are immediately unloaded when the last reference to them is lost, so the shared_ptr should be stored if the library is actually needed.
Changes applied:
* Moved utility files to /util/.
* Removed unused #includes.
* Removed unused ::ffmpeg::tools function.
* Removed unused variables.
* Fixed missing parentheses in the version macro.
* Fixed missing override on virtual function overrides and removed unnecessary virtual keyword from them.
* Disabled additional warning for ATL headers on MSVC only.
* Replaced direct printf parameters with their macro equivalent.
* Replaced C-style casts with C++-style casts.
* Applied clang-format again after an earlier change to the CMake file broke the integration for it.
The levels 5.2 and above were added in a newer FFmpeg version than OBS Studio is shipping with. Ideally we'd take these options directly from FFmpeg, but that is for a future commit.
Fixes the issue where "high444p" shows as "high", and "high" shows as "main", caused by the internal option mapping never being correct across version changes. Instead of keeping our own copy of things that has to be adjusted for each FFmpeg version, it's easier to just use the FFmpeg values and names.
* Changes the encoder name to 'streamfx-{name}' from 'streamfx--{name}' as the latter is a typo, but adds a proxy to still support the latter in bad configurations.
* Some of the warning messages have been improved in order to better support end-users, and support for the new encoder error messages has been added.
* Adds support for the is_hw argument instead of blindly relying on obs_encoder_get_caps() which actually returns the wrong values due to rerouting.
* Fixed handler-less encoders showing up in the UI outside of debug builds.
Due to an oversight in the code, the fallback code would simply fail to work as the encoder was always the zero-copy version. This resulted in weird crashes and encoders not working that used to work before.
Fixes#295
* "Quality" Minimum/Maximum is actually QP Minimum/Maximum
* Bitrate Limits is now just Limits
* Buffer Size and Quality Target have been moved into "Limits".
Proxies allow for older configurations to work fine on newer versions, without having to manually adjust the scene collection to match the new ids at all. Thanks to the migration system we can freely support any number of old versions, as long as we write migration code.
Occasionally, mostly due to other sources rebuilding their UI, an out-of-order lock freeze can be observed with Source Mirror. This is unwanted, so we need to move the freezing logic into a place where freezing shouldn't happen.
Fixes#228
Actually fixes#61
Moves the menu for StreamFX to the primary menu, so that it is not hidden under tools. This makes it clearer to the user if their installation of StreamFX is working correctly, in addition to reducing the steps necessary to use the menu.
Also the 'About StreamFX' dialog now actually shows up for every update, as expected.
Adds support for enumerations, a different way of selecting how something should behave in a shader. Enumerations rely on a continuous list of values, and will automatically detect how many values there are in the enumeration. Only non-vector types are supported as enumeration entries, and array/vector parameters can have each member set to a different enumeration value.
Furthermore suffixes now are properly assigned, and 'bool' no longer causes shaders to stop rendering. Additionally by inlining some functions and using std::string_view we can achieve a slightly better performance than before.
Using std::string_view over std::string (and const std::string&) has the advantage that we skip potential temporary std::string objects that are immediately thrown away, thus slowing down the code. It can also be implicitly cast to std::string, which makes it compatible with existing code that uses std::string.
The high priority CUDA stream causes libOBS to be at a lower priority than the tracking, which is not what we want. Instead we want tracking to be incomplete in those cases, rather than slowing down encoding and other things.
Geometry updates are also now done once per frame instead of one per tracking update, which should improve the smoothness without affecting performance too much. Additionally all tracking info is now in the 0..1 range, which drastically simplifies some math - especially with texture coordinates.
To deal with tracking and updates being asynchronous, a very simple approximation of movement velocity has been added. This is mostly wrong, but it can bridge the gap where tracking updates are slower, as the values are all filtered anyway.
Adds a new CMake option "ENABLE_PROFILING" which enables all CPU and GPU performance profiling available in StreamFX for tracking what's actually causing things to be slow.
Asynchronous rendering allows the GPU to perform work while the CPU performs other work, and is significantly faster than lockstep immediate rendering. By reusing existing render targets we can see a performance improvement of up to 500%, while still doing the same things.
The new libOBS API allows us to directly access the underlying API instead of having to mess around in memory. By using it we can avoid crashing in case the compiler for it is different, or in case the actual back end structure changes.
Additionally the mostly unimplemented and unused options have also been removed, which streamlines the use of this class even further and reduces both shader and code complexity.
Finally by optimizing the use of the internal render target we can achieve a speed up of up to 3000% over the old way, allowing for many more mipmapped filters.
Q_INIT_RESOURCE and Q_CLEANUP_RESOURCE can't be called from within a namespace and instead have to be in outside of the namespace, so by moving them into small inline functions we can fulfill this restriction.
Related: #192#155
