model not working on multi person

NVIDIA-AI-IOT / trt_pose

Real-time pose estimation accelerated with NVIDIA TensorRT

MIT License

957 stars 290 forks source link

model not working on multi person #129

Open aniketzz opened 3 years ago

aniketzz commented 3 years ago

When I run the model on two-person in 1 frame, it's working fine. But when more than 2 person comes in the frame, the 2nd person's keypoints gets distributed between the 2nd and 3rd or nth person.

I am unable to pinpoint the issue. The model is not able to decide which coordinate belongs to which person in the case of more than 2 people.

I have attached the screenshots of the results for referance. 2021-05-25

nireshd commented 2 years ago

Yeah, i have the same problem. model only detects two people

HZNUJeffreyRen commented 2 years ago

I have the same problem, Did you solve it？

aniketzz commented 2 years ago

I did the same thing in Deepstream. Current it is set to detect only 2 people by default. Changing max_num_objects = 2 to max_num_objects = 100 did the trick.