Instructions:

1. Download the IPython Notebook from the link below.
2. Upload the notebook onto a SageMaker Notebook Instance OR Install necessary packages on the desired compute resource to use the notebook.
3. Bring in the input images for Table Detection onto the SageMaker Notebook Instance OR on the desired compute resource.
4. Image file size<4 mb. We are supporting .png and .jpg image formats .
5. Following are the Types of tables which will be accurately detected by Model. a. The Table should be fully or partially bounded by the lines on all sides. b. The Columns inside the table must have proper spacing or separating lines between them. c. In case of Multi-Table Image Document, the tables shouldn't be adjacent. d. The Table should contain more than two columns if column-headers are not present. e. Image should have proper separation between free-text and tabular regions for accurate predictions.
6. Follow the instructions in the IPython Notebook for rest of the setup and consuming the service.

   Input

   Supported content types: application/json The images need JSON serialized to be fed to the model. Code can be found in jupyter notebook.

Output

Content type: application/json

Please use the following snippet to save the json content into a image file (details in jupyter notebook ):

def prediction_wrapper(prediction):
    p_json_parse = json.loads(prediction)
    return p_json_parse['processedImage'] 

table_image = np.array(prediction_wrapper(prediction))
cv2.imwrite("image_process.png",table_image)

Invoking endpoint

AWS CLI Command

You can invoke endpoint using AWS CLI:

aws sagemaker-runtime invoke-endpoint --endpoint-name "endpoint-name" --body fileb://input.json --content-type application/json --accept application/json out.json

Resources

• Sample input images
• Sample Notebook
• Sample Outputs