Known issues, tips & FAQ¶

A loose collection of notes — promote items into the main guide as they firm up.

Cameras & connection¶

Camera won't connect — check the order (battery → power on → USB), that the cable is data-capable, and that the serial number is in go2kin_config.json. (🚧 TODO: further diagnostics — does it enumerate in Windows? Can you reach the derived IP?)
Popup warning when connecting ("camera checkbox not available") — harmless; a leftover from a removed Recording-tab checkbox. Fix planned.
Camera rejects a Resolution/FPS option — the camera returns an error and a popup lists the actually available options; pick one of those.
60 Hz regions / other camera models — the Resolution/FPS lists are hardcoded for the Hero 12 with 50 Hz anti-flicker (Australia). For 60 Hz regions the options are typically 24/30/60/120/240 — edit the combo values in code/GUI/main_window.py (global_fps_var, global_res_var). For other models, run the settings discovery tool first.

WARN status — the offsets from the two claps disagree by more than one frame. Re-record with louder, sharper, well-separated claps inside the first 3 seconds.
Claps not detected — (🚧 TODO: common causes — background noise, claps too quiet or too late — and fixes.)
High frame rates — at 100+ fps, camera-to-source distance differences cause sub-frame errors; set the sound source position in the Recording tab (requires a loaded calibration).
During extrinsic calibration — sound-delay compensation can't apply (camera poses don't exist yet); clap near the centre of the camera volume so distance differences largely cancel.
Sync fails with "ffmpeg trim failed … received no packets" / "error in an external library" / "Driver does not support the required nvenc API version" — at high resolution / frame rate (e.g. 2.7K @ 200 fps) the trim step needs ffmpeg's hevc_nvenc (GPU) encoder. The default conda ffmpeg has no NVENC; install a full NVENC build and (if you get the nvenc API version message) one matching your NVIDIA driver. See Installation → NVENC ffmpeg.

Printed board doesn't match the configured size — always measure the printed squares and enter the measured value in the board config.
Poor extrinsic result — (🚧 TODO: symptoms and fixes; quality metrics and pass/fail heuristics are on the roadmap.)
Re-defining the lab axes — Set Origin can be re-run after loading a saved calibration.

Painfully slow — you're probably running on CPU. Check onnxruntime-gpu is installed (not onnxruntime) and that the log reports CUDA.
Phantom "person 1" with default 1.75 m / 70 kg — stale files from a previous run in pose-3d/ and kinematics/ get globbed into the new run. Workaround: delete those two folders inside the trial's processed/ directory before re-running. An automatic fix is planned.
Validation fails — check that synced videos exist, a calibration was saved, and the participant has height and mass entered.

Keep digital zoom fixed after intrinsic calibration — changing it invalidates intrinsics.
Hypersmooth and HindSight are turned off on connect deliberately (frame warping / battery) — don't re-enable them.

🚧 TODO: accumulate lab wisdom here — battery life expectations, cable routing, board handling and storage…

Can I use WiFi instead of USB? No — by design, for reliability in the lab. For WiFi-based GoPro control, see Go2Rep.
More than 4 cameras? Not currently supported.
Can I use it outdoors? Not the design target — audio-clap synchronisation is unlikely to be reliable outside.
Other GoPro models? Any model with the HTTP API (Hero 9+) should work — run the settings discovery tool and adjust the Resolution/FPS lists to match.

🚧 TODO: add questions as they come up.