Repository navigation
Troubleshooting and Diagnostics
Begin with the active route, not the visible symptom.
A flat picture, a late picture and a beautifully timed wrong picture have different causes, however united they may feel.
SyLC has several source, decoder and presentation routes. Good diagnosis starts by recording which one is active: source type, detected stereo layout, decoder, display mode, colour path and — for 2D→3D — provider/model/grid and depth-map age.
Before changing settings, note:
- SyLC build/version;
- Windows version and GPU model/driver;
- source container, codec, resolution, bit depth and stereo representation;
- whether the source is a local file, ISO, BDMV folder or physical disc;
- on-screen format badge and selected presentation mode;
- whether the problem affects the embedded view, detached output, both projectors or Quest;
- whether it survives pause, seek and a second known-good file.
The rotating application log is stored at:
%LOCALAPPDATA%\SyLC3DPlayer\logs\sylc-player.log
Up to five rotated 8 MB logs are retained. Set SYLC_LOG_LEVEL=DEBUG before launch when a normal INFO log does not expose enough detail.
Never attach copyrighted media to a public issue. A short synthetic/reproducible sample you created or are allowed to share is ideal; otherwise provide probe metadata and the smallest relevant log interval.
| Symptom | First question | Likely area |
|---|---|---|
| MVC file plays flat | Does the badge say MultiView 3D? | MVC detection, missing dependent view or native decoder fallback |
| Depth is reversed | Does swapping eye order fix it? | View ordering, not convergence |
| Film stutters only from disc | Does the same title play from ISO? | Optical reads or reader contention |
| HDR is grey/washed out | Is Windows HDR enabled on the output display? | Transfer/output path |
| Audio is consistently early/late | Is the offset stable? | Output-chain latency; use [ / ] trim |
| 2D→3D is unavailable | Are model packs installed? | Model manifest/files or source eligibility |
| 2D→3D shimmers or lags | What are provider, grid, map rate and map age? | Inference cadence, flow/stabilization or cuts |
| Quest sees the PC but receives nothing | Did discovery succeed while TCP failed? | Firewall, Wi-Fi route or receiver connection |
| Export menu is disabled | Is the source a supported full-resolution stereo family? | Eligibility or missing tools |
- Read the format badge.
- Confirm the source truly contains both views. An MVC rip made without the dependent view is ordinary H.264 afterward, regardless of its filename.
- For MKV, check that MVC signalling/track data survived the remux.
- For BDMV, open the disc/folder rather than a guessed base
.m2ts. - Look for a native MVC initialization failure followed by an mpv/base-view fallback in the log.
- Test a known-good MVC sample to separate source authoring from local runtime problems.
Do not force a packed-SBS presentation onto a single base view and call the duplicated image stereo. Two copies of Paris are still one camera.
First swap eye order. If that corrects front/back placement, stop there.
If order is correct but the picture is tiring:
- reduce 2D→3D Depth Strength;
- bring Convergence toward a comfortable screen-plane balance;
- compare authored stereo with synthesis to isolate display configuration;
- verify the display has not independently swapped eyes;
- test at the actual viewing distance rather than nose-to-panel engineering distance.
Convergence moves the scene relative to the screen. Strength changes the disparity budget. Neither fixes incorrect left/right identity.
Physical media should be tested as a storage system:
- Clean the disc.
- Close indexing, ripping and playback tools that may read the same drive.
- Avoid concurrent thumbnail/probe jobs.
- Archive to ISO with SHA-256 enabled.
- Replay the same timestamps from the ISO.
If corruption is repeatable from the disc and absent from a clean ISO, the optical path is the leading suspect. If it remains at exactly the same decoded frame in the ISO, preserve the log and source metadata for decoder investigation.
See Blu-ray, ISO and disc archiving for the reasoning behind the single-reader policy.
- Put the detached output on the intended HDR display.
- Enable HDR for that display in Windows.
- Confirm the display reports its HDR mode.
- Test known PQ/BT.2020 content.
- Temporarily compare
SYLC_NATIVE_HDR=1andSYLC_NATIVE_HDR=0to identify the selected branch. - Disable remote desktop, capture filters and display-side dynamic processing during the test.
- Record source primaries, transfer, matrix, range and bit depth from the log/probe.
The correct technical background is in HDR and native rendering.
Check in this order:
- Open 2D→3D → Depth models….
- Confirm at least the Small pack is completely installed.
- Check that the selected preset is enabled; presets with no compatible fixed-shape graph are greyed out.
- Confirm
onnxruntime.dll,DirectML.dllandmodels/MANIFEST.jsonexist beside the application. - Read the log for SHA-256 mismatch, missing graph, grid mismatch or provider initialization failure.
- Disable optional TensorRT and retry DirectML to separate acceleration from synthesis itself.
The player should degrade to 2D when synthesis fails. If playback also stops, report that separately: it crosses an intended failure boundary.
The useful numbers answer different questions:
| Field | Meaning |
|---|---|
| provider / model / grid | Which engine and fixed-shape graph are active |
| depth maps/s | How often new geometry is delivered, not the video frame rate |
| map age | How old the geometry is at presentation |
infer_ms |
Neural-network inference time |
flow_ms |
Motion transport cost |
stab_ms |
Temporal fusion/stabilizer cost |
cycle_ms |
Complete worker cycle |
inwait_ms |
Time waiting for a new submitted input |
reswait_ms |
Time waiting for the result consumer/resource handoff |
pass_ms |
Renderer feed-pass period |
taplat_ms |
GPU capture-to-mailbox latency |
A slow infer_ms is compute. A large inwait_ms with fast inference means the engine is hungry, not slow. The famous case where 4K outran 1080p came from a GPU readback copy that was issued but not submitted promptly; the harder file accidentally flushed the queue. Numbers are occasionally rude enough to contradict intuition, which is why they are invited.
UDP discovery succeeded, but the Quest could not open the TCP media session. Check Windows Firewall, client isolation on the access point and that both devices are on the same reachable network.
Confirm USB debugging/ADB authorization, the receiver app, and the port-forward setup used by the sender. Replugging may invalidate the transient ADB route; the receiver should reconnect, but the PC must still see the device.
Collect the five-second pressure telemetry: receiver backlog, sender/TCP pressure, in-flight work, dropped access units, recovery-IDR requests and active bitrate. One number alone rarely identifies congestion.
Check eligibility first:
- MVC, MV-HEVC, Full-SBS, Full-TAB, or active 2D→3D;
- no half-resolution SBS/TAB;
- x265 and MP4Box present in the packaged tools tree;
- ffmpeg and ffprobe discoverable;
- even per-eye dimensions suitable for 4:2:0;
- enough free space for temporary and staged output.
If validation refuses the final file, keep the log. The refusal is a success of the safety system: it prevented a plausible-looking one-eye or partial output from replacing the destination.
Include:
- exact build identifier;
- steps from launch to failure;
- expected and observed result;
- source family and metadata, without copyrighted content;
- affected output surfaces;
- the focused log interval, ideally with DEBUG enabled;
- whether a known-good control file works;
- for performance, a stable measurement window rather than one screenshot of one frame.
Avoid leading with a theory. “Depth pauses every 42 ms while inwait_ms rises” is evidence. “DirectML hates my GPU” is a promising novel title.
Start
Watch & preserve
Under the hood