adobe/antialiased-cnns
pip install antialiased-cnns to improve stability and accuracy
Implements shift-invariant CNNs by inserting `BlurPool` layers that perform low-pass filtering before downsampling, replacing strided convolutions and pooling operations. Compatible with PyTorch architectures (ResNet, VGG, DenseNet, MobileNet, etc.), allowing antialiasing to be retrofitted into existing pretrained models via weight transfer or applied to custom architectures with minimal code changes.
1,681 stars and 17,044 monthly downloads. No commits in the last 6 months. Available on PyPI.
Stars
1,681
Forks
201
Language
Python
License
—
Category
Last pushed
Apr 08, 2024
Monthly downloads
17,044
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/adobe/antialiased-cnns"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Related frameworks
HelmchenLabSoftware/Cascade
Calibrated inference of spiking from calcium ΔF/F data using deep networks
lucidrains/RIM-pytorch
Implementation of Recurrent Independent Mechanisms in Pytorch
orchardbirds/bokbokbok
Custom Loss Functions and Evaluation Metrics for XGBoost and LightGBM
KaiyangZhou/pytorch-center-loss
Pytorch implementation of Center Loss
kimhc6028/relational-networks
Pytorch implementation of "A simple neural network module for relational reasoning" (Relational Networks)