https://github.com/finnickniu/torchvison_object_detection

https://github.com/finnickniu/torchvison_object_detection

Last synced: 16 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/finnickniu/torchvison_object_detection
• Owner: finnickniu
• Created: 2020-09-23T06:30:57.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2020-09-23T06:33:43.000Z (over 4 years ago)
• Last Synced: 2024-10-30T08:37:31.687Z (2 months ago)
• Language: Python
• Size: 38.1 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: readme.md

Awesome Lists containing this project

README

        
# Torchvision Object Detection

## Install environment

1. Build conda environment

    ```

        conda create -n demo python=3.6 -y

        conda activate demo

    ```

2. Install pytorch

    ```

        conda install pytorch cudatoolkit=10.0 torchvision -c pytorch

    

    ```

3. Install cocoapi

   ```

        pip install cython

        pip install git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI

    ```

4. Install requirement dependencies

    ```

        pip install -r requirements.txt

    ```

## Run model

1. Regarding image tagging, you may need to train un-labeled image. You could use this online tagging tool (https://app.supervise.ly) to annotate image. It is easy to use, just check their turtorial.

2. Config your model

    Currently, the API only supports MaskRCNN and FasterRCNN. You need to create a work_dir firstly. 

    Seondly, copy the config.json to your work_dir.

    

    Thirdly, modify the parameters in the config.json.

    You can choose "mask_rcnn" or "faster_rcnn" as your detection engine.

3. Train your model

    ```

        python train.py

    ```

4. Visualize yout model

    ```

        tensorboard --logdir=path to work_dir

    ```

5. Test your model

    ```

        python test.py --model_path path --cuda_device cuda:1 --video_path path --score_thr 0.6

    ```