Min-Entropy Latent Model for Weakly Supervised Object Detection
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
data/models
layers
utils
README.md
Run_MELM.sh

README.md

Prerequisites

  • Linux (tested on ubuntu 14.04LTS)
  • NVIDIA GPU + CUDA CuDNN
  • Torch7

Getting started

  1. Install the dependencies

    luarocks install hdf5 matio protobuf rapidjson loadcaffe xml
  2. Download dataset, proposals and ImageNet pre-trained model

    Download VOC2007 from: http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtrainval_06-Nov-2007.tar http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar

    Download proposals from: https://dl.dropboxusercontent.com/s/orrt7o6bp6ae0tc/selective_search_data.tgz

    Download VGGF from: http://www.robots.ox.ac.uk/~vgg/software/deep_eval/releases/bvlc/VGG_CNN_F.caffemodel https://gist.githubusercontent.com/ksimonyan/a32c9063ec8e1118221a/raw/6a3b8af023bae65669a4ceccd7331a5e7767aa4e/VGG_CNN_F_deploy.prototxt

    The data folder has the following structure:

    $MELM/data/datasets/VOCdevkit_2007/
    $MELM/data/datasets/VOCdevkit_2007/VOCcode
    $MELM/data/datasets/VOCdevkit_2007/VOC2007
    $MELM/data/datasets/VOCdevkit_2007/...
    $MELM/data/datasets/proposals/
    $MELM/data/models/
    $MELM/data/results/
  3. Install functions

    cd ./MELM
    export DIR=$(pwd)   
    
    cd $DIR/utils/c-cuda-functions
    sh install.sh
    
    cd $DIR/layers
    luarocks make
  4. Train and test

    cd $DIR
    sh Run_MELM.sh 0 VOC2007 VGGF SSW 0.1 None melm

Acknowledgements

This work would not have been possible without prior work: Vadim Kantorov's contextlocnet, Spyros Gidaris's LocNet, Sergey Zagoruyko's loadcaffe, Facebook FAIR's fbnn/Optim.lua.