Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Motivation
The Python converter code supports writing boolean weights, but not reading them. This causes problems when reading TFJS models from Python.
Furthermore, the Python converter fails to read scalar string weights. Numeric scalar weights are read just fine, but string scalars cause an error.
Implementation
np.bool
has been added to the list of supported input dtypes.The deserialization code for string arrays has been fixed so the size calculation always returns a valid
int
.Details
Scalars are stored in the weight manifest with an empty shape ([]), which is technically correct to reflect the nature of a scalar.
The string deserialization uses
np.prod()
to calculate the total number of items. Unfortunately,np.prod()
returns1.0
when given an empty iterable (see numpy docs under Notes).This value cannot be used with
range()
later, which leads to an error in situations where a weight manifest contains string scalars (e.g. model SSD OpenImages v4).This change is