I’m trying to get image crops using craft-text-detector in python.

Using craft-text-detector==0.4.3 torch==1.11.0 torchvision==0.12.0 numpy==1.23.0

Code:

# import craft functions
from craft_text_detector import (
    Craft,
    read_image,
    load_craftnet_model,
    load_refinenet_model,
    get_prediction,
    export_detected_regions,
    export_extra_results,
    empty_cuda_cache
    )

# set image path and export folder directory
output_dir = '/tmp/advance-outputs/'

# construct full path to the image
image_path = 'sample-images/image.jpg'
         
refine_net = load_refinenet_model(cuda=False)
craft_net = load_craftnet_model(cuda=False)
        
craft = Craft(output_dir=output_dir, crop_type="poly", cuda=False)  # Create the Craft instance here

# apply craft text detection to the current image and export detected regions
prediction_result = craft.detect_text(image_path)

# read image
image = read_image(image_path)

# load models
        
# perform prediction
prediction_result = get_prediction(
 image=image,
 craft_net=craft_net,
 refine_net=refine_net,
 text_threshold=0.7,
 link_threshold=0.4,
 low_text=0.4,
 cuda=False,
 long_size=1280
 )

# export detected text regions
exported_file_paths = export_detected_regions(
image=image,
regions=prediction_result["boxes"],
output_dir=output_dir,
rectify=True
)

# export heatmap, detection points, box visualization
export_extra_results(
 image=image,
 regions=prediction_result["boxes"],
 heatmaps=prediction_result["heatmaps"],
 output_dir=output_dir
)

# unload models from gpu
empty_cuda_cache()

Above code is working fine and saving image-crops in /tmp/advance-outputs.

But above code is taking image_path, I want to give it PIL Image. For this I have change image_path into PIL Image:

from PIL import Image
from craft_text_detector import (
    Craft,
    read_image,
    load_craftnet_model,
    load_refinenet_model,
    get_prediction,
    export_detected_regions,
    export_extra_results,
    empty_cuda_cache
    )


output_dir = '/tmp/advance-outputs/'

image_path = 'sample-images/image.jpg'

pil_image = Image.open(image_path)
         
refine_net = load_refinenet_model(cuda=False)
craft_net = load_craftnet_model(cuda=False)
        
craft = Craft(output_dir=output_dir, crop_type="poly", cuda=False)  

        
prediction_result = get_prediction(
 image=pil_image,
 craft_net=craft_net,
 refine_net=refine_net,
 text_threshold=0.7,
 link_threshold=0.4,
 low_text=0.4,
 cuda=False,
 long_size=1280
 )

exported_file_paths = export_detected_regions(
image=pil_image,
regions=prediction_result["boxes"],
output_dir=output_dir,
rectify=True
)

export_extra_results(
 image=pil_image,
 regions=prediction_result["boxes"],
 heatmaps=prediction_result["heatmaps"],
 output_dir=output_dir
)

empty_cuda_cache()
`

But this code is giving me error:

$ python3 text_detection_advanced.py 
Traceback (most recent call last):
  File "../pythoncode/text_detection_advanced.py", line 339, in <module>
    prediction_result = get_prediction(
  File "../site-packages/craft_text_detector/predict.py", line 47, in get_prediction
    image = image_utils.read_image(image)
  File "../site-packages/craft_text_detector/image_utils.py", line 28, in read_image
    return img
UnboundLocalError: local variable 'img' referenced before assignment

Guide me where I’m doing wrong, and how can I surpass this error.