[Docs] The issue with using DBNet for inferring curved text instances.

[GZ-Metal-Cell]

Branch

master branch https://mmocr.readthedocs.io/en/latest/

📚 The doc issue

I am using the DBNet trained on the total-text dataset to infer curved text instances:

from mmocr.apis import TextDetInferencer
inferencer = TextDetInferencer(model='dbnet_resnet18_fpnc_1200e_totaltext')
print(inferencer('demo/img144.jpg', show=True))

However, it can only output quadrilateral text boxes. How can I make it generate polygonal text boxes to accurately describe curved text instances?

{'predictions': [{'polygons': [[385.1835820124699, 1612.049594381581, 452.2935673640324, 1456.1367141060207, 929.6251690204326, 1661.7132853632388, 862.5151836688701, 1817.6261656387992], [949.715346116286, 1612.426505047342, 1113.102571927584, 1426.9068636686904, 1189.6717012845552, 1494.379940530528, 1026.284475473257, 1679.8995819091797], [357.7371626633864, 830.1354798026706, 1210.7975510817307, 813.8510473500127, 1219.4141517052283, 1265.4869295203166, 366.3537409855769, 1281.7713619729745]], 'scores': [0.8527029156684875, 0.8840392827987671, 0.7951122522354126]}], 'visualization': []}

Suggest a potential alternative/fix

No response