File Uploader File to OpenCV imread

🎈 Using Streamlit
1

How can I pass image file from file_uploader() to opencv’s imread()?

1 Like
2

file_uploader does not store the uploaded file on the webserver but keeps it in memory. It returns a ‘stream’ class variable to access it. I came across this situation as well but honestly I dont have THE answer yet. Fact seems to be that some python modules dont care whether you throw a pathname or a stream at them , but many other do. To me the python ‘stream’ classes appear as a relatively relatively complex matter. You may try to give imread the stream but if this doesnt work you may save the stream to a file and then read like usual. However it might be difficult with streamlit to find the right point in time to delete the quasi temporary file. I dont know if the webserver does this by it itself, does it? Same issue with files you create for downloading. AT the moment I am completely ignoring this issue on my streamlit apps because there are only very few users that work with them.
code fragment to save the file:

fs = st.file_uploader('upload a file')
if fs is not None:
    with open(fs.name,'wb') as f:
        f.write(fs.read())
2 Likes
3

Thanks a lot.

4

@ksxx you are my knight in shining armor.

I attempted to read file_uploader from memory using these two functions

def create_opencv_image_from_stringio(img_stream, cv2_img_flag=0):
    img_stream.seek(0)
    img_array = np.asarray(bytearray(img_stream.read()), dtype=np.uint8)
    return cv2.imdecode(img_array, cv2_img_flag)

def get_opencv_img_from_buffer(buffer, flags):
    bytes_as_np_array = np.frombuffer(buffer.read(), dtype=np.uint8)
    return cv2.imdecode(bytes_as_np_array, flags)

but they didn’t work and I would say don’t waste your time trying to get them to work.

The main problem with this is that it would create endless image files in the server-side application if each image name is different right?

5

Turns out you can just save all images as ‘img.jpg’ and it won’t create endless files with different names.

from PIL import Image

# Saves
img = Image.open(img)
img = img.save("img.jpg")

# OpenCv Read
img = cv2.imread("img.jpg")
1 Like