jhjin / overfeat-torch

OverFeat-Torch7 Wrapper Library

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

OverFeat-Torch7 Wrapper

OverFeat is a Convolutional Network-based image classifier and feature extractor from NYU. The original can be found in the repository: https://github.com/sermanet/OverFeat This application loads weights from OverFeat and construct a network for vanilla Torch7. Torch7 and extra packages (image, nn, torchffi) should be properly installed.

Install

Run the shell script to download weights and build this library.

sh install.sh

Run demo

Run the command below. By default, the script loads a small network, and categorizes the bee.jpg image using nn backend (on CPU).

th run.lua

For example, if you want to run a big model on GPU using cudnn library in a memory-efficient manner (inplace opertor), use the command instead.

th run.lua --network big --backend cudnn --inplace

Docker

If you'd prefer a faster setup process and are familiar with Docker, you can build and run a container with all of the necessary dependencies with the following commands:

docker build -t overfeat-torch .
docker run overfeat-torch
cd /root/overfeat/
th run.lua

About

OverFeat-Torch7 Wrapper Library


Languages

Language:Lua 94.9%Language:C 2.9%Language:Shell 1.5%Language:Makefile 0.8%