wai.annotations module for managing TFRecords datasets.
The manual is available here:
https://ufdl.cms.waikato.ac.nz/wai-annotations-manual/
Reads image object-detection annotations in the TFRecords binary format
- Image Object-Detection Domain
usage: from-tf-od [-I FILENAME] [-i FILENAME] [-N FILENAME] [-n FILENAME] [--seed SEED] [--mask-threshold THRESHOLD] [--sample-stride STRIDE]
optional arguments:
-I FILENAME, --inputs-file FILENAME
Files containing lists of input files (can use glob syntax)
-i FILENAME, --input FILENAME
Input files (can use glob syntax)
-N FILENAME, --negatives-file FILENAME
Files containing lists of negative files (can use glob syntax)
-n FILENAME, --negative FILENAME
Files that have no annotations (can use glob syntax)
--seed SEED the seed to use for randomisation
--mask-threshold THRESHOLD
the threshold to use when calculating polygons from masks
--sample-stride STRIDE
the stride to use when calculating polygons from masks
Writes image object-detection annotations in the TFRecords binary format
- Image Object-Detection Domain
usage: to-tf-od [--dense] -o PATH [-p FILENAME] [-s FILENAME [FILENAME ...]] [--split-names SPLIT NAME [SPLIT NAME ...]] [--split-ratios RATIO [RATIO ...]]
optional arguments:
--dense outputs masks in the dense numerical format instead of PNG-encoded
-o PATH, --output PATH
name of output file for TFRecords
-p FILENAME, --protobuf FILENAME
for storing the label strings and IDs
-s FILENAME [FILENAME ...], --shards FILENAME [FILENAME ...]
additional shards to write to
--split-names SPLIT NAME [SPLIT NAME ...]
the names to use for the splits
--split-ratios RATIO [RATIO ...]
the ratios to use for the splits