Related Publications: Fuqin Deng, Hua Feng, Mingjian Liang, Qi Feng, Ningbo Yi, Yong Yang, Yuan Gao, Junfeng Chen, and Tin Lun Lam. "Abnormal Occupancy Grid Map Recognition using Attention Network." International Conference on Robotics and Automation (ICRA), 2022. paper link: https://arxiv.org/abs/2110.09047

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, OGMD 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