Skip to content

Virtual pooling technique for accelerating convolutional neural networks

License

Notifications You must be signed in to change notification settings

joechen24/virtual-pooling-caffe

Folders and files

NameName
Last commit message
Last commit date

Latest commit

86665f4 · Oct 12, 2019
Sep 5, 2018
Sep 5, 2018
Jul 14, 2016
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Oct 12, 2019
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Sep 5, 2018
Jan 23, 2019
Sep 5, 2018

Repository files navigation

Virtual Pooling Caffe

Implementation of virtual pooling technique to speedup convolutional neural networks in Caffe.

We provide scripts to automatically generate train_val and solver prototxt for sensitivity analysis and grouped fine-tuning (need to create tmp_solver and tmp_train_val folders under caffe/models/{net} first) For example, run python ./models/resnet/modify_net.py

Script for parsing accuracy and timing output results is provided as well (need to create {net}_results folder under caffe/ first) For example, run python ./models/resnet/parse_results.py

To do sensitivity analysis, run ./build/tools/caffe test --model=models/vgg16/train_val_lininterp_conv11_afterAct.prototxt --weights=models/vgg16/VGG_ILSVRC_16_layers.caffemodel -gpu 0 -iterations 1000 ./build/tools/caffe test --model=models/vgg16/train_val_lininterp_conv12_afterAct.prototxt --weights=models/vgg16/VGG_ILSVRC_16_layers.caffemodel -gpu 0 -iterations 1000 ./build/tools/caffe test --model=models/vgg16/train_val_lininterp_conv21_afterAct.prototxt --weights=models/vgg16/VGG_ILSVRC_16_layers.caffemodel -gpu 0 -iterations 1000 . . .

To do grouped fine-tuning, run ./build/tools/caffe train --solver=models/vgg16/solver_lininterp_finetune_round1.prototxt --weights=models/vgg16/VGG_ILSVRC_16_layers.caffemodel -gpu 0

To get timing info, run ./build/tools/caffe time --model=models/vgg16/train_val_lininterp_round1_afterAct.prototxt -gpu 0

About

Virtual pooling technique for accelerating convolutional neural networks

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published