musco-ai / musco-pytorch

MUSCO: MUlti-Stage COmpression of neural networks
BSD 3-Clause "New" or "Revised" License
71 stars 16 forks source link

how to calculate the speedup in the paper #3

Open liusiping10 opened 5 years ago

liusiping10 commented 5 years ago

when I run the compressed VGG16 model, the runtime is even longer than VGG16 before compressed. how to run the compressed models getting the same results (Table 5) in the paper https://arxiv.org/abs/1903.09973?