yangdb / melm

Min-Entropy Latent Model for Weakly Supervised Object Detection

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.