This is an implementation of Personal Belongings Detection and Recognition Module in AIR Project. The module has two main parts, object detector and instance classifier. The object detector is based on faster-rcnn.pytorch, the attention module came from CBAM and BAM, and the instance classifier consists of one fc layer.

1. Clone this repository.

   git clone https://github.com/ai4r/AIR-ObjectDetection.pytorch
   cd AIR-ObjectDetection.pytorch
   pip install -r requirements.txt

2. create a model folder

   mkdir models/InstModel

3. download and install the font file

   wget http://cdn.naver.com/naver/NanumFont/fontfiles/NanumFont_TTF_ALL.zip
   unzip NanumFont_TTF_ALL.zip -d NanumFont
   rm -f NanumFont_TTF_ALL.zip
   mv NanumFont /usr/share/fonts/
   fc-cache -f -v

4. Download the model files and move to models folder.

1. Run the demo code.

   python demo_detector.py

2. The program will load images from the test_input_image folder, then save results images to the test_output_image folder.

1. Run the demo code.

   python demo.py

2. Press key 'r' and type two information, category_name and instance_name.
3. Show the object instance until predefined number of images are captured.

1. Run the demo code.

   python demo.py

2. Press key 'd' and see the result.

See LICENSE.md