a question about making own datasets

jimmyyhwu / pose-interpreter-networks

Real-time robotic object pose estimation with deep learning

MIT License

122 stars 27 forks source link

a question about making own datasets #14

Closed dingshenglan closed 5 years ago

dingshenglan commented 5 years ago

Hello, I now mark each image in my dataset with ”labelme“ and convert the annotation.json format to coco format, but these files are one by one, how to properly merge a annotation file, just like the annotations file you provide in Oilchangedatasets?

jimmyyhwu commented 5 years ago

You would write your own script to do that. See the COCO API and data format.

I would make sure the image ID and annotation ID are unique after merging, and that categories and category IDs are the same for all images.