mmsegmentation·数据结构

PixelData

• 简单的理解为torch.tensor或者ndarray的一个封装
• 常用属性为.data,..metainfo
  data属性

gt_segmentations = PixelData(metainfo=img_meta,data=torch.randint(0, 2, (1, 4, 4)))
gt_segmentations.data

metainfo：字典类型

img_meta = dict(img_shape=(4, 4, 3),pad_shape=(4, 4, 3))

tensor([[[1, 1, 1, 0],[0, 0, 1, 1],[0, 1, 1, 1],[0, 1, 0, 0]]])

---

gt_segmentations.metainfo['img_shape']
(4, 4, 3)

• 完全支持torch.tensor的常见操作：

# 类张量的操作
gt_segmentations = PixelData(metainfo=img_meta)
gt_segmentations.data = torch.randint(0, 2, (1, 4, 4))
cuda_gt_segmentations = gt_segmentations.cuda()
cuda_gt_segmentations = gt_segmentations.to('cuda:0')

cuda_gt_segmentations=cuda_gt_segmentations.detach().cpu().numpy()
print(type(cuda_gt_segmentations.data))
<class 'numpy.ndarray'>

SegDataSample

• gt_sem_seg，pred_sem_seg和metainfo
• gt_sem_seg，pred_sem_seg 貌似只接受PixelData类型的数据

import torch
from mmengine.structures import PixelData
from mmseg.structures import SegDataSampleimg_meta = dict(img_shape=(4, 4, 3),pad_shape=(4, 4, 3))
data_sample = SegDataSample()