search

GeekAlexis / FastMOT

High-performance multiple object tracking based on YOLO, Deep SORT, and KLT 🚀
MIT License
1.15k stars 253 forks source link

model to ONNX to TRT error #110

Closed manelhalimaia closed 3 years ago

manelhalimaia commented 3 years ago

I convert this model to ONNX and then to TRT with Fast-Reid repo "https://github.com/JDAI-CV/fast-reid/ " and test the inference I got the model MSMT17 Baseline AGW(S50) from https://github.com/JDAI-CV/fast-reid/blob/master/MODEL_ZOO.md It turns for the 5 (detector_frame_skip=5) first frames and than we got this error "Traceback (most recent call last): File "app3.py", line 192, in main() File "app3.py", line 95, in main mot.step(frame) File "/home/trx/Downloads/FastMOT_version_23_04_2021/fastmot/mot.py", line 109, in step self.tracker.update(self.frame_count, detections, embeddings) File "/home/trx/Downloads/FastMOT_version_23_04_2021/fastmot/tracker.py", line 151, in update cost = self._matching_cost(confirmed, detections, embeddings) File "/home/trx/Downloads/FastMOT_version_23_04_2021/fastmot/tracker.py", line 252, in _matching_cost self.max_feat_cost, self.motion_weight) ValueError: unable to broadcast argument 2 to output array File "/home/trx/Downloads/FastMOT_version_23_04_2021/fastmot/tracker.py", line 324,"

GeekAlexis commented 3 years ago

86

manelhalimaia commented 3 years ago

Thanks GeekAlexis it works well we have change the feature dimension to 2048 ;)