How to open an image

Part 1 (2020)
1

I had written in fastai v2 (still in dev) this bit of code.

def input_fn(request_body, content_type=JPEG_CONTENT_TYPE):
    logger.info('Deserializing the input data.')
    # process an image uploaded to the endpoint
    if content_type == JPEG_CONTENT_TYPE:
        img = open_image(BytesIO(request_body))
        return img

The idea was to run then learn.predict(img).

I am getting now an error as open_image cannot be found. How can I solve this issue? What happened to open_image?

Thanks a lot!!

2

open_image is still there. Did you make sure to upgrade your fastai version? (essentially what happened is the pip version of fastai2 became the 2.0.0 version of fastai

1 Like
3

Oh! Where is it to import it @muellerzr ? I am doing from fastai.vision.all import * and getting a not defined error. My fastai is 2.0.0 so this should be fine.

1 Like
4

Ah sorry, I blanked. Now we have PILImage.create instead to open an image up.

2 Likes
5

I am running this:

# Deserialize the Invoke request body into an object we can perform prediction on
def input_fn(request_body, content_type=JPEG_CONTENT_TYPE):
    logger.info('Deserializing the input data.')
    # process an image uploaded to the endpoint
    if content_type == JPEG_CONTENT_TYPE:
        return PILImage.create(io.BytesIO(request_body))
    # process a URL submitted to the endpoint
    if content_type == JSON_CONTENT_TYPE:
        img_request = requests.get(request_body['url'], stream=True)
        return PILImage.create(io.BytesIO(img_request.content))
    raise Exception('Requested unsupported ContentType in content_type: {}'.format(content_type))

but for some reason I get the error cannot identify image file <_io.BytesIO object at 0x7fe521605e08>
I tried with Pillow==6.2.0 as well as the last version of the library (7.2.1) but did not help either. I find it very weird as this runs in my sagemaker conda_pytorch_p36 kernel but not when I deploy the pytorch fastai estimator

import io
import requests
from PIL import Image
r = requests.get(urls[0], stream=True)
im = PILImage.create(io.BytesIO(r.content))

I don’t really know what else to try so any ideas are very welcome, getting a bit desperate in here.

6

Don’t run the io.Bytes bit. PILImage can accept that. Just do r.content (if you check the source you’ll notice it’ll actually check for that)

7

True! but without io.BytesIO I was getting this error so basically this is why I added it but it did not help get around the issue though