Add "append" mode for convert_imageset. #5762

Open
wants to merge 4 commits into
from

Conversation

Projects
None yet
1 participant

tonykcn commented Jul 11, 2017

Sometimes, we need to append additional images into lmdb/leveldb, especiallly in fine-tuning stage.

If we construct all data with convert_imageset from ZERO, it's inefficient when the image count is large.

If "append" is set explicitly, new image data are appended after existing rows, no need to touch existing constructed data.

Tong Jin added some commits Jul 11, 2017

Tong Jin Add "append" mode for convert_imageset.
Sometimes, we need to append additional images into lmdb/leveldb, especiallly in fine-tuning stage.
If we construct all data with convert_imageset from ZERO, it's inefficient when the image count is large.
If "append" is set explicitly, new image data are appended after existing rows, no need to touch existing constructed data.
f4617ec
Tong Jin Add "append" mode for convert_imageset.
Sometimes, we need to append additional images into lmdb/leveldb, especiallly in fine-tuning stage.
If we construct all data with convert_imageset from ZERO, it's inefficient when the image count is large.
If "append" is set explicitly, new image data are appended after existing rows, no need to touch existing constructed data.
a2dcf4e
Tong Jin Add "append" mode for convert_imageset.
Sometimes, we need to append additional images into lmdb/leveldb, especiallly in fine-tuning stage.
If we construct all data with convert_imageset from ZERO, it's inefficient when the image count is large.
If "append" is set explicitly, new image data are appended after existing rows, no need to touch existing constructed data.
b443bd0
Tong Jin Add "append" mode for convert_imageset.
Sometimes, we need to append additional images into lmdb/leveldb, especiallly in fine-tuning stage.
If we construct all data with convert_imageset from ZERO, it's inefficient when the image count is large.
If "append" is set explicitly, new image data are appended after existing rows, no need to touch existing constructed data.
93b4066

tonykcn commented Jul 12, 2017 edited

Sometimes, we need to append additional images into lmdb/leveldb, especiallly in fine-tuning stage.
If we construct all data with convert_imageset from ZERO, it's inefficient when the image count is large.
If "append" is set explicitly, new image data are appended after existing rows, no need to touch existing constructed data.

@shelhamer , could you help review and merge this change? Any suggestion/comment will be appreciated.

Thanks in advance.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment