NetsPresso

Model Compressor

Ready-to-use toolkit

Model Compressor makes your compression process easy and fast

Available until June, 2022

Just put your model

You can experience amazing compression!

Modules-compressiontoolkit-v1.png

Benefits 

time_save.png

A ready-to-use

compression toolkit

You can compress your AI model within 5 minutes. Just upload and compress models.

network.png

Supports

all CNN architectures

You can compress any AI model based on CNN architectures.

tune.png

Optimal

compression ratio

You don’t need to worry about the method configurations. Reduce trial and error with our recommendation feature.

save-time.png

Eliminates months of

paper implementation period

You can only focus on the COMPRESSION, not wasting time implementing complicated methods.

loss.png

Minimal loss of information

You can fine-tune the compressed model after the compression.

Support

All Convolutional Neural Networks 

 Framework 

Pytorch
TensorFlow-Keras

 Key features 

Network visualization
Model profiling
Recommendation

 Supported methods​ 

Structured Pruning
  • Pruning by Index
  • Pruning by Criteria
    • L2 Norm Pruning
    • Geometric Median Pruning
    • Nuclear Norm Pruning
Filter Decomposition
  • Tucker Decomposition
  • CP Decomposition

Best practices

Check out the 40x improvement

Classification benchmark-v1.png

Best Practices for Classification

* NPTK-Simple: FD and pruning were performed once using NetsPresso Model Compressor.

* NPTK-Search: Best practice based on Nota's internal research using NetsPresso Model Compressor.

Best Practices for Object Detection

* NPTK-Simple: FD and pruning were performed once using NetsPresso Model Compressor.

Best Practices for Super Resolution

* NPTK-Simple: FD and pruning were performed once using NetsPresso Model Compressor.

* NPTK-Search: Best practice based on Nota's internal research using NetsPresso Model Compressor.