Using NetsPresso Model Compressor on a variety of well-known public datasets, this article shows users how to get outstanding compression results.
The fine-tuning procedure is necessary for each compression. It usually follows the original model's training configuration, except the learning rate. After a few batches of training, the learning rate is optimized by determining if the loss has converged or not.
All of the original and compressed models can be downloaded easily on the Model Compressor Model Zoo.
See Image Classification Results
See Object Detection Results
See Semantic Segmentation Results
See Super Resolution Results
Updated over 1 year ago