To address grid map classification problem, we construct a dataset containing 6916 (the labels including 3210 normal and 3706 abnormal )grid maps through an indoor robot vacuum cleaner. These grid maps are created with an initial size of 50m×50m. To further increase the number of training examples, we applied random rotation and offset to cropped areas of 34m×34m used as training examples. To the best of our knowledge, OGMCD is a large-scale benchmark specifically for indoor grid map classification.

1. Inside of OGMCD/python/ directory run conda create -n myenv python=3.6.
2. Activate the virtual environment by running source activate myenv
3. Install requirements from requirements.txt by running pip install -r requirements.txt

We train 400 epochs by Stochastic Gradient Descent (SGD)with the momentum of 0.9 and a weight decay of 1e-4. The learning rate starts from 0.01 and drops every 50 epochs. It takes about 10 hours for the network to converge on an NVIDIA GTX 2080Ti graphics card.

python train.py [OGMCD with train and val folders] train [path to weights file saves] -a [model name]

For example

python train.py [OGMCD-folder with train and val folders] train ./model_save/ -a se_resnet32

python test.py [OGMCD with test folders] test [path to weights file] -a [model name]

For example

python test.py [OGMCD with test folders] test se_resnet32.pth -a se_resnet32

You may download the dataset reported in the paper from Google Drive or the Baidu Netdisk

Baidu Netdisk eval code：yyvs