Compare camera hardware range vs ML model-limited range at a given resolution and disparity budget
Formulas
Z_min = f_px × B / d_max · Z_max = f_px × B / EPE · ΔZ = Z² × EPE / (f_px × B) · Far@N% = (N/100 × f_px × B) / EPE · f_px_model = f_px_native × (model_width / native_width)
Model disparity error (EPE)
This input is the model's empirical disparity error (end-point-error) in pixels — how far off its predictions actually are, on average, at the distance you care about. Lower EPE → longer far range and better accuracy for the same resolution and baseline. A benchmark's published global-average EPE isn't distance-uniform — disparity error typically grows with distance — so the honest input here is error measured at your actual target distance, not the flat dataset average. If unsure, use 1 px (conservative).
Note
Max disparity (d_max) only affects Z_min (near plane). Far range and depth error depend on f_px × B and the model's disparity error (EPE). Camera ideal ranges are manufacturer specs verified from official sources.