Padding sensitive batch construction for text recognition
Techniques for constructing padding-sensitive batches for text recognition are provided. In one technique, a first bounding box from a list of bounding boxes is added or included into a batch of zero or more bounding boxes. Each bounding box in the list of bounding boxes surrounds different detected text in a digital image. A second bounding box is identified from the list. The second bounding box is wider than the first bounding box. A difference between (1) a width of the second bounding box and (2) a particular width is determined. The particular width is based on a width of a bounding box in the batch. Based on the difference and a threshold value, it is determined whether to include the second bounding box in the batch. The batch is then input into a test recognition model.
1 . A method comprising:
including, into a batch of zero or more bounding boxes, a first bounding box from a list of bounding boxes, wherein each bounding box in the list of bounding boxes surrounds different detected text in a digital image;
identifying a second bounding box, from the list of bounding boxes, that is wider than the first bounding box;
after including the first bounding box into the batch, determining a difference between a width of the second bounding box and a particular width that is based on a width of a bounding box in the batch;
based on the difference and a threshold value, determining whether to include the second bounding box in the batch;
inputting the batch into a text recognition model;
wherein the method is performed by one or more computing devices.
2 . The method of claim 1 , further comprising:
sorting the list of bounding boxes based on width to generate a sorted list of bounding boxes.
3 . The method of claim 1 , wherein determining whether to include the second bounding box in the batch comprises:
comparing the difference to the threshold value;
including the second bounding box in the batch if the difference is less than the threshold value.
4 . The method of claim 1 , wherein determining whether to include the second bounding box in the batch comprises:
comparing the difference to the threshold value;
creating a new batch and including the second bounding box in the new batch if the difference is greater than the threshold value.
5 . The method of claim 1 , wherein the second bounding box immediately follows the first bounding box in the list.
6 . The method of claim 1 , wherein the bounding box is the first bounding box.
7 . The method of claim 1 , wherein:
prior to including the first bounding box into the batch, the batch includes one or more bounding boxes;
the particular width is based on multiple widths of multiple bounding boxes in the batch.
8 . The method of claim 7 , further comprising:
performing an aggregation operation on the multiple widths to generate the particular width.
9 . The method of claim 8 , wherein the aggregation operation is one of mean, median, maximum, minimum, or percentile.
10 . The method of claim 1 , wherein the list of bounding boxes is a plurality of bounding boxes, further comprising:
resizing a subset of the plurality of bounding boxes to have a particular height.
11 . One or more non-transitory storage media storing instructions which, when executed by one or more computing devices, cause:
including, into a batch of zero or more bounding boxes, a first bounding box from a list of bounding boxes, wherein each bounding box in the list of bounding boxes surrounds different detected text in a digital image;
identifying a second bounding box, from the list of bounding boxes, that is wider than the first bounding box;
after including the first bounding box into the batch, determining a difference between a width of the second bounding box and a particular width that is based on a width of a bounding box in the batch;
based on the difference and a threshold value, determining whether to include the second bounding box in the batch;
inputting the batch into a text recognition model.
12 . The storage media of claim 11 , wherein the instructions, when executed by the one or more computing devices, further cause:
sorting the list of bounding boxes based on width to generate a sorted list of bounding boxes.
13 . The storage media of claim 11 , wherein determining whether to include the second bounding box in the batch comprises:
comparing the difference to the threshold value;
including the second bounding box in the batch if the difference is less than the threshold value.
14 . The storage media of claim 11 , wherein determining whether to include the second bounding box in the batch comprises:
comparing the difference to the threshold value;
creating a new batch and including the second bounding box in the new batch if the difference is greater than the threshold value.
15 . The storage media of claim 11 , wherein the second bounding box immediately follows the first bounding box in the list.
16 . The storage media of claim 11 , wherein the bounding box is the first bounding box.
17 . The storage media of claim 11 , wherein:
prior to including the first bounding box into the batch, the batch includes one or more bounding boxes;
the particular width is based on multiple widths of multiple bounding boxes in the batch.
18 . The storage media of claim 17 , wherein the instructions, when executed by the one or more computing devices, further cause:
performing an aggregation operation on the multiple widths to generate the particular width.
19 . The storage media of claim 18 , wherein the aggregation operation is one of mean, median, maximum, minimum, or percentile.
20 . The storage media of claim 11 , wherein the list of bounding boxes is a plurality of bounding boxes, wherein the instructions, when executed by the one or more computing devices, further cause:
resizing a subset of the plurality of bounding boxes to have a particular height.