METHOD & LIMITS · UPDATED OCTOBER 14, 2026
Know what the video can show.
DeadliftMy is an experimental browser tool for reviewing one conventional deadlift set. Its tracking accuracy, original phone-video compatibility and usefulness against existing tools have not yet been established.
Inputs and timestamps
The current technical baseline is an H.264 MP4 at 30 or 60 frames per second, no longer than 45 seconds, no larger than 100 MiB, and up to 1920 pixels on the long side / 1080 on the short side. This is a supported test envelope, not a guarantee that every phone recording will load. WebCodecs and OffscreenCanvas support are required.
The parser uses MP4 presentation timestamps rather than inferring elapsed time from a screen recording. Variable timing, unsupported transforms and complex editing metadata are rejected for quantitative review. Slow-motion exports can still change the meaning of time; confirm that the source was recorded and exported at normal speed.
Tracking visible pixels
You select the actual camera-side plate center and a reference circle. A fixed-template correlation search follows image content on reduced-resolution frames. This is an image-matching heuristic, not a trained barbell recognition system. Low matching scores, ambiguity and search-boundary hits create gaps. It does not replace a missing plate with a wrist point or interpolate an unknown segment.
Patterns, occlusion, blur, small plates and changing perspective can cause loss or undetected drift. Inspect the entire selected path. Set a new point at a problem frame and retrack; confirm the boundaries again after editing.
Comparisons and units
Mark the first sustained lift-off and the end of upward motion manually. Two upward phases play from lift-off on a common elapsed-time scale; the shorter side holds at its endpoint. Record whether each rep is dead-stop, touch-and-go or unknown.
Paths are aligned at their starting point and use the same reference-circle diameter as a relative image scale. These are 2D plate-diameter units, not centimetres or metres per second. No absolute speed, power, 1RM, injury risk, technique score or load prescription is provided. Differences in repetition style, camera position and frame selection limit interpretation.
Current evidence
Development checks include ten automated tests and a 4-second generated H.264 motion test. They demonstrate parts of the software pipeline; they do not establish real training accuracy. Tests on an authorized 20-clip set, physical Android/iPhone devices and eight users comparing their existing workflow are still pending. No real training example or expert endorsement is presented on this site.
Saving and clearing
Review edits live in page memory. Refreshing loses them. PNG cards are generated locally and downloaded only when you request it; choose an image-free card to omit original video frames. Clearing the review stops processing and releases the app’s references; it does not erase your original file or a previously downloaded PNG.