tianzhi0549 / FCOS

FCOS: Fully Convolutional One-Stage Object Detection (ICCV'19)
https://arxiv.org/abs/1904.01355
Other
3.28k stars 630 forks source link

How to define labels when training on own datasets? #174

Closed milliema closed 5 years ago

milliema commented 5 years ago

As there are many forms of COCO labels, for object detection and semantic segmentation the annotations are different. What are the exact form of the labels if I use my own dataset to train? Mine is like below: { "id": 645, "image_id": 40, "category_id": 2, "segmentation": [ [ 5289, 663, 5295, 663, 5295, 681, 5289, 681 ] ], "area": 108, "bbox": [ 5289, 663, 6, 18 ], "iscrowd": 0 },