This is C++ version of FairMOT using TensorRT in Windows. The default model has been replaced with dlav0 to ignore the dependency of DCN. I have trained this model in the CrowdHuman dataset. If you would like to use the original DCN model, please refer to another repository FairMOT_TensorRT.
export onnx model file in FairMOT by adding the following code in line 479 in "/src/lib/models/networks/pose_dla_conv.py", and put it to the folder "models"
z = {}
for head in self.heads:
z[head] = self.__getattr__(head)(y[-1])
hm = z["hm"]
wh = z["wh"]
reg = z["reg"]
hm = F.sigmoid(hm)
hm_pool = F.max_pool2d(hm, kernel_size=3, stride=1, padding=1)
id_feature = z['id']
id_feature = F.normalize(id_feature, dim=1)
id_feature = id_feature.permute(0, 2, 3, 1).contiguous() #switch id dim
return [hm, wh, reg, hm_pool, id_feature]
- Kalman Filter is borrowed from DeepSort, [deep_sort] https://github.com/apennisi/deep_sort
- [FairMOT](https://github.com/ifzhang/FairMOT)