open-rppg 0.0.6

Open rPPG

open-rppg

An easy-to-use rPPG inference toolbox

Installation

Python >= 3.9

pip install open-rppg

Import and Use

import rppg
model  = rppg.Model()
result = model.process_video("your_video.mp4")

Result Example

{'hr': 100.45004500450045,        # Heart Rate (FFT Method)
 'SQI': 0.8665749931341046,       # Signal Quality
 'hrv':{
    'bpm': 103.6194862200475,     # Heart Rate (Peak Method)
    'ibi': 579.0416666666666,     # Inter-Beat Interval
    'sdnn': 54.76628055757589,    # Standard Deviation of NN intervals
    'sdsd': 30.674133962201175,   # Standard Deviation of Successive Differences
    'rmssd': 46.25344260031846,   # Root Mean Square of Successive Differences
    'pnn20': 0.5714285714285714,  # Proportion of NN50 > 20ms
    'pnn50': 0.2857142857142857,  # Proportion of NN50 > 50ms
    'hr_mad': 8.333333333333314,  # Heart Rate Median Absolute Deviation
    'sd1': 29.276576197229755,    # Short-term variability
    'sd2': 59.75143144804733,     # Long-term variability
    's': 5495.642490576809,       # Poincar矇 Plot Area
    'sd1/sd2': 0.489972800445545, # SD1/SD2 Ratio
    'breathingrate': 0.21607605877268798,
    'VLF': 0.09521664913596516,   # Very Low Frequency Power
    'TP': 2.056694418632364,      # Total Power
    'HF': 1.2267116642737315,     # High Frequency Power
    'LF': 0.7347661052226675,     # Low Frequency Power
    'LF/HF': 0.5989721355243509   # LF/HF Ratio
  },
 'latency': 0.0}                  # Real-Time Latency

Real-Time Mode

import time
model = rppg.Model()

with model.video_capture(0):          # Connect to your webcam
    while True:
        result = model.hr(start=-15)  # Get heart rate from last 15 seconds
        if result:
            print(f"Heart Rate: {result['hr']} BPM")
        time.sleep(1)

Real-Time Frame Preview

for frame, box in model.preview:     # Current RGB frame and detection box
    x, y  = box                      
    face  = frame[x[0]:x[1], y[0]:y[1]]

Get BVP Wave

bvp, ts        = model.bvp()         # BVP with timestampes
raw_bvp, ts    = model.bvp(raw=True) # Unfiltered BVP

Time Slice

now       = model.now                      # Video duration or current time
bvp, ts   = model.bvp(start=10, end=20)    # BVP slice from 10 to 20 seconds
bvp, ts   = model.bvp(start=-15)           # The last 15-second slice
hr        = model.hr(start=-15)            # HR of the last 15 seconds 

Model Selection

print(rppg.supported_models) # ['ME-chunk.rlap', 'ME-flow.rlap', .......]
model = rppg.Model('RhythmMamba.rlap') # RhythmMamba trained on rlap

Pretrained Models

Model	Training Set	Description	Paper
ME-chunk	PURE RLAP	rPPG based on state-space model	2025
ME-flow	PURE RLAP	ME in low-latency real-time mode	2025
PhysMamba	PURE RLAP	Mamba with fast-slow network	2024
RhythmMamba	PURE RLAP	Mamba with 1D FFT	2025
PhysFormer	PURE RLAP	Transformer with central diff conv	2022
TSCAN	PURE RLAP	Conv attention with temporal shift	2020
EfficientPhys	PURE RLAP	TSCAN with self attention	2022
PhysNet	PURE RLAP	3D CNN encoder-decoder network	2019

Use CUDA

Install JAX with CUDA (Linux only).

pip install jax[cuda]