feat(player): improve enhancement quality and preserve video geometry - #755
Merged
Conversation
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_dbf43106-a99b-459d-a619-8b47f4e2d645) |
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
Contributor
Documentation previewThe documentation preview has been deployed for this pull request. |
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_c4500fd3-b418-4b30-84dd-62e5480e79db) |
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_ab667cc2-fa70-4822-9bd4-45008d21aab5) |
stackia
added a commit
that referenced
this pull request
Sep 9, 2026
Fix the brief black flash after the first picture when picture enhancement or deinterlacing is enabled. This fixes a startup regression introduced in #755. The first video-frame callback can arrive while `readyState` is still `HAVE_METADATA`, so `primeCanvas()` skips drawing. The callback now renders that decoded frame when startup left the texture ring empty, instead of exposing an empty canvas until the next frame. Successfully primed frames still avoid a duplicate upload and filter pass. Validation: five focused regression tests cover enhancement, deinterlacing, both enabled, already-primed startup, and raw-video fallback. The three affected cases fail before the fix. All 48 frontend tests, TypeScript, changed-file Biome checks, and the production frontend build pass. Six Chrome/WebGL playback cases using generated fixtures, including progressive and 1080i H.264/MP2 MSE playback, show no black frames after the first picture and no GL errors. The low-frame-rate fixture reproduces the black flash before the fix. E2E lint and both collection entry points also pass (626 tests collected; daemon tests were not run for this frontend-only change). Fixes #760. <!-- CURSOR_SUMMARY --> --- > [!NOTE] > **Low Risk** > Narrow change to first-frame scheduling in the video render loop, with regression tests; no auth, data, or API surface changes. > > **Overview** > Fixes a **startup black flash** when picture enhancement or deinterlacing is on: the first `requestVideoFrameCallback` can fire before `readyState` reaches `HAVE_CURRENT_DATA`, so `primeCanvas()` skips and the canvas stays empty until the next frame. > > **`scheduleFrame`** now calls `processFrame` when the texture ring is still empty (`textures.length === 0`), not only when the loop was already running (`wasRunning`). Frames that were already primed still skip a second upload/filter pass. > > Adds **`renderer.test.ts`** with Vitest coverage: first-frame present for enhancement/deinterlace combos, no double upload when startup is already primed, and no GL path when both processing features are off (mocked presenters/GPU). > > <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit 99c97d4. Configure [here](https://www.cursor.com/dashboard/bugbot).</sup> <!-- /CURSOR_SUMMARY -->
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Improve denoising and display-size scaling for SD and 1080p/i playback, and prevent the picture changing proportions when enhancement is toggled on anamorphic channels such as Shanxi Satellite TV.
Validation: 43 frontend tests, TypeScript checking, changed-file Biome checks, and the production frontend build passed. Real WebGL checks on Chromium covered 32 combinations of native, smaller, and upscaled output, independent axis scaling, Y-flipping, caller-bound framebuffers, contrast and saturation, and returning from upscale to native size; both RGBA16F and RGBA8 denoising history paths passed. Six playback cases verified the expected shader stages for 1080p/i at 1920 × 1080, 2560 × 1440, and 3840 × 2160, plus 24 anamorphic geometry checks covering paused toggles and deinterlacing. Live preview checks on CCTV-1 at all three output sizes and on Shanxi Satellite TV confirmed the expected scale uniforms and contrast/saturation values; paired GPU replays of captured frames had zero pixel error. No GL errors were observed.
Note
Medium Risk
Large changes to the real-time WebGL video path (shaders, temporal state, and sizing) affect every enhanced playback session, though failures fall back to passthrough with explicit resets on seeks.
Overview
Replaces the spatial mosquito-nr filter with a new TemporalDenoiser at source resolution: motion-adaptive history blending, variance clipping on cuts, and original luma packed in alpha for downstream sharpening. History resets on seeks, stage changes, and timeline discontinuities.
FSR path is reworked: EASU runs only when the canvas is larger than the source; native-size output still gets denoise + RCAS with stronger, noise-aware luma sharpening (using denoiser/EASU alpha), scale-aware edge gating, and RGB10_A2 intermediates. Bit-hack reciprocals are replaced with guarded exact math.
The renderer keys uploads and the render gate off rVFC decoded dimensions (
metadata.width/height), not SAR-scaledvideoWidth/videoHeight, so anamorphic rasters keep correct aspect when enhancement is toggled. Frame textures use pre-allocated RGB8 storage plustexSubImage2Dfor ANGLE/Metal stability;createProgramcleans up shaders on link/compile failure.Pipeline wiring moves resize/discontinuity handling into the renderer (seeking, emptied, lightweight rVFC when stopped) and caps upscale targets with
MAX_TEXTURE_SIZE.Reviewed by Cursor Bugbot for commit 9327ca6. Configure here.