LMDBDataset

LMDBDataset

zcls supports LMDB to speed up data set reading

How to create LMDBDataset

See pnno

tip: before use pnno to create LMDB file, you can use tools/zoom.py to resize image

How to Use

modify config_file like this

DATASET:
  NAME: 'LMDBDataset'
  TRAIN_ROOT: '/path/to/train_lmdb_path'
  TEST_ROOT: '/path/to/test_lmdb_path'
  TOP_K: (1, 5)

Note: TRAIN_ROOT and TEST_ROOT should fill in the corresponding path

For ImageNet

zcls specifically implements the use of Imagenet. Modify config_file like this

DATASET:
  NAME: 'LMDBImageNet'
  TRAIN_ROOT: '/path/to/train_lmdb_path'
  TEST_ROOT: '/path/to/test_lmdb_path'
  TOP_K: (1, 5)

Related references

• What’s the best way to load large data?