This is a Dockerfile that builds a docker image of PVANET

1.This image can only be built inside the IronYun domain, otherwise the caffemodel for PVANET will be missing

2.Currently the subordinate class of detection result are meaningless, it's only output examples

Build the image:

1.Clone this repository

git clone https://github.com/GBJim/pvanet-docker.git

2.Build the image with nvidia-docker

cd pvanet-docker
nvidia-docker build -t <IMAGE_NAME_YOU_LIKE> .

Run the demo:

1.Start the image with nvidia-docker

nvidia-docker run -ti <IMAGE_NAME_YOU_LIKE>

2.Once you are attached to the image, move to the pvanet directory

cd ~/pva-faster-rcnn/

3.Run the demo.py

python tools/demo.py

The output will be in JSON format. A list containing info of multiple objects, each object also contains the information of subordinate object class

[
{'ymax': 156.07154846191406, 'score': 0.84535301, 'xmax': 183.18292236328125, 'xmin': 133.978515625, 'ymin': 111.8216781616211, 'class': u'person', 'sub': {'score': 0.84535301, 'class': 'sedan/SUV'}}, 

{'ymax': 301.84246826171875, 'score': 0.99305266, 'xmax': 426.51141357421875, 'xmin': 89.23002624511719, 'ymin': 44.30303955078125, 'class': u'bus', 'sub': {'score': 0.99305266, 'class': 'van'}}
]

4.If you want to run the demo.py with a specific GPU.

python tools/demo.py --gpu 1

Check the output classes:

See the datasets/config.py for the specs of main and subordinate classes.