SEMI-LINES

This is the official PyTorch implementation of our semi supervised method for line segement detection.

Pretrained Weights

There are some pretrained weights for the two datasets and some splits.

├── ./exp/
    ├── finnswoods/
        ├── supervised/
        └── semisup/
    └── wireframe/
        ├── supervised/
        └── semisup/
     

Dataset

• Finnwood: git
• Wireframe: git
• Spruce A & B: Download

Annotations from Finnwood are already in data/FinnForest/.

Please modify your dataset path in configuration files.

├── data/
    ├── FinnForest/
        ├── rgb/
        ├── annos_train_finnwoods.json
        └── annos_val_finnwoods.json
    ├── sam_segs/
        ├── skrylle_frames/
        ├── snoge_frames2/
        ├── annos_snoge.json
        └── annos_skrylle_new.json
    └── Wireframe/
        ├── images/
        ├── train.json
        └── valid.json

Usage

Docker

Pull the Docker Image

Pull the Docker image from Docker Hub using the following command:

docker pull johannaengman/semi-lines:latest

This will fetch the Docker image used to run this code.

Running the Container with Mounted Data

To run the container and mount your local data, use the following command:

docker run -it --gpus all -v /path/to/your/data:/home2/johannae/semi-lines/UniMatch johannaengman/semi-lines:latest

Explanation:

• /path/to/your/data: Replace with the directory of your code and data.
• -it: Runs the container interactively.
• -v /path/to/your/data:: Mounts your local directory /path/to/your/data to the /home2/johannae/semi-lines/UniMatch directory inside the container. Replace /path/to/your/data with the actual path to your dataset.

Running the Training Script

Once the container is running and your data is mounted, execute the script inside the container:

sh scripts/train.sh  <num_gpu> <port>

This will run the semi-supervised method with pretraind weights from the supervised method. To run the supervised method in order to obtain the weights, run scripts/train_supervised.sh

Running the Inference Script

If you want to run inference on the best model obtained during training run:

sh scripts/inference.sh  <num_gpu> <port>

In the inference script you can change what method you want to run and with which test dataset.

Citation

If you find this project useful, please consider citing: