Questions about exporting coco format annocation file

[zrazerx]

I want to use MMOCR to train text detection and text recognition in a specific scene. I tried to use Labelbee to label my dataset , my setting like this

then I exported COCO format, here is the annotation file:

{"images":[{"id":1,"file_name":"C:\\Users\\XXX\\Desktop\\1\\20220208_1_01.JPG","width":810,"height":810,"valid":true,"rotate":0}],
"annotations":[{"image_id":1,"id":1,"iscrowd":0,"segmentation":[[534.9400299850075,184.58770614692656,683.095952023988,182.15892053973016,684.3103448275863,233.1634182908546,535.1458772539607,232.08197636829112]],"area":7321.901405358607,"bbox":[534.9400299850075,182.15892053973016,149.37031484257875,51.00449775112443],"category_id":0,"textAttribute":"text content","order":1}],
"categories":[{"id":0,"name":"","supercategory":""}]}

So, I want to know whether this COCO-like format annotation can be directly used for MMOCR or MMsegmentation？

[Kerwin-L]

Sorry, The COCO-like format is not directly used for MMOCR or MMsegmentation. You can do it currently by yourself if you want to use MMOCR.

I talk to the owner of MMOCR today. They will unify the format in the future.

If you want to use MMsegmentation, you can export by Mask.

But sorry for that, there are a bug.🥲 #33. I will fix it in the next few days. I will use another way like python to solve it