From the API it reads
data ( str , bytes , BytesIO , numpy.ndarray , or file opened with )
But it doesn’t seem to work if I load the video as a numpy array. Am I using this right?
import numpy as np
import streamlit as st
t, h, w, c = 10, 20, 20, 3
video = np.random.normal(0, 1, t*h*w*c).reshape((t, h, w, c))
st.video(video)
Hi @komodovaran,
Yes, Streamlit accepts numpy arrays for video, audio, and image data. However, those arrays must contain the necessary headers and image information already (streamlit doesn’t construct them for you).
http://www.fileformat.info/format/tiff/corion.htm
There is a “Todo” in the Streamlit code about making an API through which Streamlit might construct headers on the user’s behalf. Where that might fit on the product roadmap might be a ways off though. 
Is this header an issue on github? I’m interested in tracking this issue.
There wasn’t one, surprisingly, so I made one. 
@komodovaran – I’m still researching how you would create the headers for a TIFF video; meanwhile check out this gist showing how to generate TIFF images starting from nparrays.
I tried the first snippet and it worked. pip install tifffile and then (reposting here):
from tifffile import imsave
import numpy as np
# create data
d = np.ndarray(shape=(10,20), dtype=np.float32) # also supports 64bit but ImageJ does not
d[()] = np.arange(200).reshape(10, 20)
# save 32bit float (== single) tiff
imsave('test.tif', d) #, description="hohoho")Both are writing to a file, then loading them on.
Is there a way to directly display the numpy data as image to save the i/o operations?
This will be especially useful if i want to display the image with a slider that adjusts the numpy data in real-time