why read url images using cv2.imread()?

[adda1221]

Hi, i find that the images is read from url directly using opencv:
image = cv2.imread(image_path)
and the image_path is from:
image_path, caption = sample['url'], sample['caption']

the requestion is, can the url be read by opencv and does this method slow down training?

[Jiaming Han]

Sorry for the confusion. The open-sourced dataset is modified from our internal code that reading images from s3 server. If you are using a dataset in your local machines, the url should be the file path.

[adda1221]

Sorry for the confusion. The open-sourced dataset is modified from our internal code that reading images from s3 server. If you are using a dataset in your local machines, the url should be the file path.

thanks for reply, got it! can i read the data directly from s3? because the memory of the local machine is small.

[Jiaming Han]

I'm afraid of not. You need to add the s3 loading function based on your s3 system's instruction.