1.1.0

Reliability, tunability and preview-performance release. Makes a high-bitrate RTSP source (e.g. a 1080p60 stream) hold its frame rate without the packet-loss corruption ("green bands") that aggressive low-latency UDP defaults caused, and surfaces what the engine is actually doing.

Added

• Two more FFmpeg engine options in Settings → Network, both applied on the next connection:
  • UDP socket buffer (buffer_size) — enlarges the kernel receive buffer so the packet bursts of a high-bitrate 1080p frame don't overflow it (the silent-drop cause of green bands). Shown in KB; 0 leaves libav's default.
  • Max delay (max_delay) — demuxer reorder delay, previously hard-coded to 0, now tunable (ms).
• Inline "?" help on every engine option: hover shows a one-line explanation, click opens the guide at that option's section. Backed by a single shared table (EngineOptionHelp) so the tooltips and the guide never drift.
• In-app engine guide (EngineGuideForm, menu Guide): one scrollable section per option with a fuller explanation and a link to the relevant FFmpeg documentation. Localized in all four languages (IT/EN/ES/DE).
• Live measured received bitrate. The SDP rarely advertises a bitrate on live RTSP (so the probe showed n/a); it is now computed from the actual demuxed packet sizes over a ~1s window and shown live in the Connection panel — VCam_GetFfmpegProducerBitrate (producer) / GetPreviewBitrate (preview).
• Active engine settings shown in the Connection table: transport preference, hardware decode, socket timeout, RTP reorder, UDP buffer, max delay and latency cap, mirrored live from the current settings.

Changed

• New "real-time but robust on average" defaults. reorder_queue_size 8 → 512 and UDP buffer_size 0 → 2 MB, so UDP survives a high-bitrate stream (no more green bands) while max_delay=0 + nobuffer + the latency cap keep it real-time. Existing settings.json values are preserved — the new defaults only apply to fresh installs.
• Preview rendering decoupled from decoding. The decode thread's sink now only packs the latest NV12 frame into a staging buffer and signals; a dedicated render thread does the NV12→BGRA convert and the GDI blit. Previously that ~15–18 ms convert+blit ran on the decode thread and throttled the receive rate below the source's fps, building a backlog that the latency cap then dropped as progressive frame loss. The GDI stretch also moved from HALFTONE to the far cheaper COLORONCOLOR.
• Diagnostics tables no longer flicker. Each refresh is batched with BeginUpdate/EndUpdate (one repaint per tick instead of one per cell), cells are only rewritten when their value actually changed, and both ListViews are double-buffered.

Fixed

• Bitrate row showed n/a on live RTSP sources; it now reflects the measured throughput.

1.0.1

Maintenance release: a small UI-string fix and documentation brought in line with the FFmpeg receive engine.

Fixed

• Italian "Start" button label showed Avvia!! (stray exclamation marks) instead of Avvia.

Documentation

• README (EN + IT) corrected for the FFmpeg receive path. The receiver is no longer Media Foundation, so the guidance built around MF's packetization quirks is gone:
  • Removed the "force a single H.264 slice per frame / sliced-threads=0 is non-negotiable" warning — FFmpeg reassembles multi-slice streams correctly (as ffplay always did). The encoder recipe stays, reframed as general low-latency advice.
  • Dropped the "RX ≫ real framerate ⇒ multi-slice H.264" diagnostic tell and the matching troubleshooting entry, both of which described the removed MF symptom.
  • Rewrote the diagnostics section for the current UI: two panels (Connection / Stats) instead of one bar, the full current field set (State, Engine, Decode, RX, Render, Duplicates, Dropped, Processing, Drift), and the fact that the stats reflect the Frame Server once the virtual camera is running (not preview-only).
  • Documented the settings added in 1.0.0 (selectable RTSP transport with live active-transport display, FFmpeg engine tuning, frame-counter overlay).

1.0.0

First stable release. The receive pipeline is now a single, tunable FFmpeg user-space engine feeding the Media Foundation virtual camera through shared memory.

Added

• FFmpeg receive path (now the only one). The app decodes the RTSP stream in user space with FFmpeg (RTCamNative/FfmpegRtspSource — d3d11va hardware decode when available, else software, with low-latency demux options and a resync-to-live latency cap) and streams NV12 frames to the Frame Server through a frame shared-memory channel (Shared/VCamFrameChannel.h, triple-buffered, seqlock, created by the Frame Server via FrameChannelReader and written by the app via FrameChannelWriter). The Frame Server reads those frames in MediaStream::RequestSample and shows the synthetic frame when the app producer's heartbeat goes stale. FFmpeg (LGPL, dynamic, decode-only) is provided by vcpkg manifest mode with vcpkg as a git submodule at external/vcpkg (version pinned via the baseline); its DLLs are app-locally deployed into the build output and bundled by the MSI — no manual setup for the end user.
• The live preview is now FFmpeg-based too (RTCamNative/FfmpegPreviewPlayer): the same decode core renders NV12→BGRA into the WinForms panel with a double-buffered GDI path. It inherits the latency cap, so a fast/misbehaving source no longer drifts the way the old Media Foundation preview did.
• Live Frame-Server stats channel (cross-process shared memory, Shared/VCamStats.h): RX/render/declined fps and copy cost, surfaced in the app's diagnostics bar while the virtual camera is running. The GPU-decode indicator comes from the app producer (VCam_IsFfmpegProducerHardware), shown as FFmpeg HW/FFmpeg SW.
• Selectable RTSP transport, with a live view of the one actually in use. Settings → Network offers Auto (UDP, TCP fallback) / UDP only / TCP only. In Auto the decode loop drives the fallback itself — it opens with an explicit UDP transport first and, if an attempt connects but carries no frame within the socket timeout, flips to TCP on the next attempt — so the connection panel shows the real transport (UDP/TCP) rather than just the request. Exposed to the UI via VCam_GetActiveTransport (producer) and GetActiveTransport (preview).
• FFmpeg fine-tuning in Settings (process-wide, applied on the next connection): hardware-decode on/off (force software decode when a d3d11va driver misbehaves), socket timeout, RTP reorder-buffer depth, and the resync-to-live latency-cap threshold. Persisted in Settings and pushed to the native engine through VirtualCameraWrapper.ApplyEngineSettings.
• Diagnostic frame-counter overlay (Settings, off by default): burns a per-frame counter into every delivered frame (real and synthetic) so the actual consumer-side frame rate is visible on the video itself.
• Settings dialog strings localized in all four supported languages (Italian, English, Spanish, German).

Removed

• The Media Foundation receive engine, entirely. The Frame Server no longer opens the RTSP source itself: RtspSessionManager and RtspReaderCallback (in-process IMFSourceReader + DXVA GPU zero-copy) were removed, as was the Media Foundation preview player (VideoPlayer/VideoReaderCbk/CSourceOpenMonitor). The engine selector (Settings → "Motore di ricezione", Settings.VideoEngine, MF_VCAM_ENGINE attribute, VCamConfig.engine) is gone — there is a single FFmpeg receive path. The virtual camera itself stays Media Foundation; only the source of its pixels changed. As a consequence the camera is live only while the app is running (the decode happens in the app process); this was previously the FFmpeg-mode trade-off and is now the only behaviour.

Changed

• deploy_vcam.ps1 now copies the built DLL to C:\Projects\RTVirtualCamera (override with -DeployDir) before registering it, instead of registering directly from the repo build output under C:\Users\... — fixes E_ACCESSDENIED on IMFVirtualCamera::Start (Frame Server runs as Local Service, which cannot access the user profile).
• VCamMediaSource::SetupCameraSettings clamps an implausible RTSP-declared frame rate (>60fps) to 30fps, instead of trusting it for MediaStream pacing.
• MediaStream::RequestSample tracks how many deliveries re-serve the same frame as the previous one (declinedFrames, count only) instead of leaving renderedFrames indistinguishable from genuinely new frames. (An attempt at actually declining these re-serves via MF_E_SAMPLEALLOCATOR_EMPTY was reverted — it destabilized the Frame Server's own retry timing on real hardware.)
• Default RTSP transport is now UDP with automatic TCP fallback (previously hardcoded TCP); a small RTP reorder buffer keeps UDP usable without shredding the picture on minor packet reordering.
• Frame Server sample delivery reworked to an async two-queue design, decoupling the RTSP receive cadence from the consumer's RequestSample polling.
• Trace output is off by default in production. ETW (WINTRACE) and the app file log (DebugLog) are gated behind runtime environment variables (RTVCAM_TRACE machine-wide for the Frame Server, RTVCAM_LOG for the app), read once and cached; when disabled they pay no formatting or file-I/O cost and WINTRACE skips EventRegister entirely.
• MSI (Setup) no longer builds in Debug configurations (Release only).

---

0.2.0

Added

• Spanish and German added to the language selector (alongside Italian and English).
• New application icon (webcam glyph on navy rounded-square background), used by the app and by the installer's Add/Remove Programs entry.
• Installer: MajorUpgrade support for clean in-place upgrades, and a Start Menu shortcut.

Changed

• Migrated RTVirtualCamera to .NET 10 (LTS), SDK-style project, self-contained deployment — the MSI no longer requires a pre-installed .NET runtime on the target machine.
• Minimum supported Windows version set to Build 22000 (Windows 11 21H2), matching the MFCreateVirtualCamera API requirement.
• About dialog redesigned.
• Settings dialog redesigned; language selector changed from radio buttons to a dropdown.

Removed

• PerformanceTests1 project (placeholder MF-infrastructure tests, never wired into the solution).

---

0.1.0

Added

• GPU zero-copy frame path: when the H.264/H.265 decoder runs on the GPU (DXVA),
  frames are copied with a single CopySubresourceRegion — no CPU copy, no allocation.
• RTSP auto-reconnection: 60 attempts at 1-second intervals after a stream drop.
• WiX installer (RT-VirtualCam-Setup.msi) with version derived from git tags.
• Settings form with localization support (IT/EN).
• Auto-start option that opens the stream on application launch without blocking the UI.
• About page.
• Performance test project for MediaStream::CopyRtspFrame.
• CI workflow (GitHub Actions) with MSI artifact upload.

Changed

• Renamed the Media Foundation bridge project from MFPipeline to RTCamNative.
• Preview stats bar moved from bottom to top of the window.
• User data (settings, logs) moved to %LOCALAPPDATA%.
• RTSP preview latency reduced.

Fixed

• Bugs A/B/C/D/F/G/H identified in code review (see CLAUDE.md for details).
• LNK2001 QISearch in RTCamNative Release|x64.
• PAUSED→RUNNING transition (SetStreamState) returning E_POINTER.
• Memory leak on Stop/Start cycle.