Hi, I wanted to generate LMDB for new dataset. I noticed for TextVQA, the currently available LMDB stores training and validation set features using keys with prefix 'train/...' ; whereas, for test set, it uses keys with prefix "test/..."
I wonder during training and evaluation, how the code differentiates the validation set from training set since they share the same prefix for their keys. Plus, defaults.yaml, does not contain location to IMDB files such as "imdb_val.npy" etc.
soonchangAI
commented
2 years ago
❓ Questions and Help
Hi, I wanted to generate LMDB for new dataset. I noticed for TextVQA, the currently available LMDB stores training and validation set features using keys with prefix 'train/...' ; whereas, for test set, it uses keys with prefix "test/..."
I wonder during training and evaluation, how the code differentiates the validation set from training set since they share the same prefix for their keys. Plus,
defaults.yaml, does not contain location to IMDB files such as "imdb_val.npy" etc.