./deep_sort/deep/checkpoint
For the ./deep_sort/deep/checkpoint the link is : https://pan.baidu.com/s/1ZI_UVUsPC9NKPFmjyF4MiQ And the password is : 1234
or you can try this link to download the file : https://drive.google.com/file/d/1QYQ6I3j8iwXWwnV6uKbKosRkzCpUpBdv/view?usp=sharing
The python version of these project is 3.7
After download these weights file and put it to the correct place, we should run this command to setup our environment.
pip install -r requirement.txt
use the command :
python3 person_count.py --video_path 'your video path'
if you have this problem ：
ImportError: libGL.so.1: cannot open shared object file: No such file or directory
Then use this command to install opencv-python-headless :
pip install opencv-python-headless
After install all the dependences of the project. We should go to our virtual environment path:
cd your_virtual_environment_path/lib/lib/python3.7/site-packages/torch/nn/modules/upsampling.py
And change the upsampling.py file. In line 152:
def forward(self, input: Tensor) -> Tensor:
# return F.interpolate(input, self.size, self.scale_factor, self.mode, self.align_corners, recompute_scale_factor=self.recompute_scale_factor)
return F.interpolate(input, self.size, self.scale_factor, self.mode, self.align_corners)
And if your environment is x86 environment, please open the reid_clothes.py file to the file 265:
ln = self.clo_net1.getLayerNames()
# ln = [ln[i - 1] for i in self.clo_net1.getUnconnectedOutLayers()]
ln = [ln[i[0] - 1] for i in self.clo_net1.getUnconnectedOutLayers()]
Finally, we only should use these command to run the project.
python3 person_count.py --video_path 'your video path'
And we improved the accuracy of the reid. And the new file is reid_improved.py you can use this command to use this file:
python3 reid_improved.py --video_path 'your video path'
And to accommodate reid detection in a wider range of situations. We add self.new_ID_Threshold in line 105. You can adjust this parameter, which ranges from 0 to 1, depending on how densely populated the scene is, and turn it down when the scene is densely populated. When the scene is sparse with people, you can turn the parameter up.
For the camera as input option, we will add it in a later optimization.