Result inconsistent

[riyajatar93]

i am doing inference with below code , it return detection and recognized text.

infer = MMOCRInferencer(det='dbnetpp', rec='svtr-small')

is there any way i can get output in json format like the one
ocr.readtext gives.

also order of text detected on input image is not consistent , i am unable to sort them properly . any method or suggestion to solve this
thanks

[gaotongxiao]

The output structure has been significantly refactored and aligned with the global design of OpenMMLab, and therefore you might need a custom postprocessor to achieve this effect.

Currently there is no universal box/text sorting algorithm, but MMOCR has one implementation that you might be interested in:

mmocr/mmocr/utils/bbox_utils.py

Line 131 in b18a09b

def stitch_boxes_into_lines(boxes, max_x_dist=10, min_y_overlap_ratio=0.8):