Launching today
RepStandard
Computer vision counts your reps in real time
75 followers
Computer vision counts your reps in real time
75 followers
Most fitness apps make you manually log sets or need a wearable to guess your effort. RepStandard's camera does the counting for you - squats, push-ups, sit-ups, and plank, tracked in real time via on-device pose detection (nothing leaves your phone). It also builds an adaptive daily program that scales with your progress, and turns consistency into a game: ranks, XP, streaks, and shareable certificates.
RepStandard
@edvinasapp Congrats on the launch 🚀
I really like that everything runs on device. For a fitness app that uses a camera, privacy is a huge trust factor, and keeping everything local makes a big difference.
I'm curious, after testing with real users, which exercise was the hardest to detect accurately? Push ups, squats, planks, or something else? It feels like solving those edge cases is where the real magic happens.
Wishing you both an awesome launch! 💪🔥
How does the camera actually handle things like messy form or switching between exercises quickly? I get that on-device processing is great for privacy, but curious how reliable the rep counting ends up feeling in a real sweaty home workout.
RepStandard
@emrahtwgn Thanks for asking.
Rep counting accuracy was actually one of the biggest technical challenges.
The hardest part wasn't counting reps, but avoiding false counts from movements like getting into the push-up or sit-up position, while also not missing real reps.
We spent countless hours filming ourselves and refining the recognition logic. After a lot of iteration, we've reached an accuracy we're really happy with. Still things can get a bit messy if there's a lot of movement happening in the background behind the user.
RepStandard
@tan_z_tan For this app we're staying focused on bodyweight workouts.
I think load recognition would be very difficult, mainly because every gym uses different plates and equipment.
However the current app recognizes barbell back squat reps pretty well. Try it. Also thrusters, front squats or anything else that involves squatting works too.
the on-device pose counting is such a smart move, especially for privacy. one thing i'd love to see is a quick audio cue or short rest timer that kicks in between sets automatically when the camera detects you've stopped, so i'm not constantly staring at the screen to check when to go again.
RepStandard
@yeldem50906 Hey. The audio cues you mentioned are already implemented, so you don't need to look at the screen during the workout. We even have multiple AI voice types for the cues.
Building on Emrah's reliability question with the part nobody's hit yet: on-device pose tracking for a full session is a thermal budget problem as much as an ML one. A phone running the camera + a pose model continuously for 30 minutes heats up and throttles, and that's exactly when inference gets flaky — so "accurate" in a 2-minute demo and "accurate at minute 25 of a sweaty circuit" can be different apps. Do you duty-cycle the model between reps or downsample frames to stay under the thermal ceiling, or is it running full-rate the whole set? And when the phone's propped against a water bottle at a bad angle, does it miscount silently or tell the user to reposition? The silent miscount is the one thing that'd make me stop trusting the number.
Real-time rep counting sounds useful for solo training too. How does the model handle exercises where body parts are partially occluded or viewed from different camera angles?
Tried the squat counter in my living room and it actually caught my rep count without me touching the screen. The XP and streak thing feels a bit cheesy but it's weirdly motivating to keep showing up.