Machine learning enabled document deskewing

A computer-implemented method, comprising:

• determining, based at least on a first image of a first document, a plurality of text bounding boxes present in the first document;
• applying a machine learning model trained to determine, based at least on a first plurality of coordinates defining the plurality of text bounding box present in the first document, a second plurality of coordinates of a first document bounding box enclosing one or more text bounding boxes present in the first document, wherein the machine learning model is applied to a threshold quantity of text bounding boxes present in the first document, wherein in response to the first document including less than the threshold quantity of text bounding boxes, duplicating at least one of the plurality of text bounding boxes present in the first document to generate the threshold quantity of text bounding boxes;
• determining, based at least on the second plurality of coordinates of the first document bounding box, one or more transformations;
• and
• deskewing the first image of the first document by at least applying, to the first image of the first document, the one or more transformations.

A non-transitory computer readable medium storing instructions, which when executed by at least one data processor, result in operations comprising:

• determining, based at least on a first image of a first document, a plurality of text bounding boxes present in the first document;
• applying a machine learning model trained to determine, based at least on a first plurality of coordinates defining the plurality of text bounding box present in the first document, a second plurality of coordinates of a first document bounding box enclosing one or more text bounding boxes present in the first document, wherein the machine learning model is applied to a threshold quantity of text bounding boxes present in the first document, wherein in response to the first document including less than the threshold quantity of text bounding boxes, duplicating at least one of the plurality of text bounding boxes present in the first document to generate the threshold quantity of text bounding boxes;
• determining, based at least on the second plurality of coordinates of the first document bounding box, one or more transformations;
• and
• deskewing the first image of the first document by at least applying, to the first image of the first document, the one or more transformations.