jcollomosse
commented
10 months ago We have instead opted to sideload the models on first use so withdraw the limit raise request
Closed jcollomosse closed 10 months ago
jcollomosse
commented
10 months ago We have instead opted to sideload the models on first use so withdraw the limit raise request
Project URL
https://pypi.org/project/trustmark/
Does this project already exist?
New Limit
600
Update issue title
Which indexes
PyPI
About the project
trustmark is an open source image watermarking library, for embedding invisible identifiers in images.
TrustMark has been developed as part of the Adobe-led Content Authenticity Initiative (CAI) coalition comprising over 2000 industry, academic and public sector members. We are releasing TrustMark under MIT licence as an open source watermarking package to promote adoption of CAI and its related open provenance standard (C2PA) developed under the Linux Foundation.
TrustMark is AI based method and requires distribution of large model files each between 40-90Mb. See the models subfolder.
There are 3 variants of TrustMark (C,B,Q) and we barely can fit the smallest of these (C) within the 100Mb current release v0.2.0.
Even then, we cannot fit the model for rewatermarking (C) which is a desirable feature.
We bundle no additional packages in our project, or example data.
In an attempt to reduce the size we tried to compress the models but they do not zip well (deflate only about 5%). We prefer not to rely on external download of the models due to dependencies and reliability issues this would introduce.
We already reduced the package size to only the (C) variant to release v0.2.0 but it is desirable also to release the remover for (C) and release also the (B,Q) variants as described in the associated technical paper below:
TrustMark: https://arxiv.org/abs/2311.18297
Reasons for the request
We ask to raise the limit to 600Mb from the current default limit of 100Mb
There are 3 variants of TrustMark (C,B,Q) and we barely can fit the smallest of these (C) within the 100Mb current release v0.2.0.
We wish to add the re-watermarking capability to (C) - requiring extra 150Mb. The code is in the lib already but we had to drop the model size.
Since we intend to release also high quality (B,Q) variants we require a total of approximately 600Mb.
Please could we increase the limit on this project to 600Mb in order add the AI model blobs to enable re-watermarking functionality and the higher quality variants (B,Q) of TrustMark.
Code of Conduct