Perform OCR multiple images at the same time?

[harsh-miv]

I am using easyocr on a GPU, but the overall time taken for my dataset is large and my GPU are not being fully utilized, is there a way to group together images and using GPU to its max extent?

[rkcosmos]

try adjusting batch_size argument (see documentation )

[alexander-soare]

@rkcosmos thanks for picking up on this. It's not clear to me how to pass in a batch of images corresponding to batch_size. I have tried with a list bit I get the misleading NameError: name 'LOGGER' is not defined error.

[sangeethkumar1997]

@rkcosmos result_batch = reader.readtext(i,detail = 0,batch_size=64)

i is of shape (64, 256, 256, 3) , But it is expecting np array of len(img.shape) less than or equal to 3.

UnboundLocalError: local variable 'img' referenced before assignment

[rkcosmos]

Input image has to be one image (not array of images). This batch_size means EasyOCR will create batch of several text boxes to be recognized at the same time internally. Batch of several input images is not supported.

[SamSamhuns]

I did create some edits on the scripts to load a batch of input images for the detector as well (inference speed should go up in GPU mode). If there is enough interest, I can create a PR for this

[sangeethkumar1997]

@SamSamhuns That would be great!

[SamSamhuns]

#458 PR has been created for the batched image inference for text detection as well, @rkcosmos please have a look