region_classification

Region Classification

In this task we predict the semantic category for each region within a house.

Requirements

• Install Torch on a machine with CUDA GPU
• Install cuDNN v5
• download pretrained resnet model

  wget https://d2j0dndfm35trm.cloudfront.net/resnet-50.t7
  

pre-trained models

Training	Model name
Single Image	snapshot-single.t7
Panorama Image	snapshot-pano.t7

To download the models:

python download_mp.py  --task_data regoin_classification_data 

Training

imgList=[path training image list] labelList=[path training image list] gpu=1 snapshotsFolder=[output path] th train.lua

Example:

imgList=./data/train_room_pano_image.txt labelList=./data/train_room_pano_label.txt gpu=1 snapshotsFolder=./pano th train.lua

The snapshot models will be saved in [output path].

Testing

imgList=[path training image list]  labelList=[path training image list] gpu=1 modelPath=[path to model] th test.lua