Can streamlit display 60 fps video from frames?

[kevinlinxc]

I’ve been using Streamlit for a real time dashboard, and while data has been fast, video has been really slow. When I say video, I mean frames, because this would be reading from a camera or server providing images. Another way of framing this question is “can st.image be run 60 times per second?”

For a while, I blamed factors like network for the slow video speed, but I tested it today and it seems like even the most bare bones st.image display loop only gets around 14 fps for a 480p video. (It’s 70ms step time, 1000/70 ~=14). Code to recreate:

480.mp4

import streamlit as st
from timeit import default_timer as timer
import cv2

vid_area = st.empty()
data_area = st.empty()
step_time = st.empty()

old_time = timer()
cap = cv2.VideoCapture("480.mp4")
while cap.isOpened():
    time = timer()
    step_time.write(f"Step time: {time - old_time:.3f}")
    old_time = time
    ret, image = cap.read()
    vid_area.image(image, channels='BGR')

I know it’s not opencv’s fault because imshowing it with opencv is very fast:

cap = cv2.VideoCapture("480.mp4")
while cap.isOpened():
    ret, frame = cap.read()
    cv2.imshow("hi", frame)
    cv2.waitKey(1)

I recognize that st.image is probably not built for this purpose, but I’m wondering what my optoins are as I really enjoy working with streamlit but I need the faster fps to use it for my use case.

[randyzwitch]

Hi @kevinlinxc -

In this case, I think you’ve answered your own question!

Given Python in general as an interpreted language, and the Streamlit execution model, I’m not sure there is an obvious solution to boost frame rates much more than what you are seeing. Maybe someone else with more experience can stop by to suggest something else to try.

Best,
Randy