How to run the 256x256 model?

[lucyd42]

I was able to successfully run the 224x224 model after a couple days of trying. I want to run the 256x256 model provided in the readme. When I download it and try to use it with demo.ipynb, it says "keyError : 'engine' (full output below).

I suspect this is beacause the new model might have some different json keys/parameters becaus it is different in structure, so my question is how do I run the 256x256 model using the sample code, what all changes will I have to make?

Console Output :

Reading TensorRT model....... Traceback (most recent call last): File "/home/lucyd/Work/trackAi/v1.0/trt_pose/tasks/human_pose/live_demo.py", line 35, in <module> model_trt.load_state_dict(torch.load(OPTIMIZED_MODEL)) File "/home/lucyd/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 830, in load_state_dict load(self) File "/home/lucyd/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 825, in load state_dict, prefix, local_metadata, True, missing_keys, unexpected_keys, error_msgs) File "/usr/local/lib/python3.6/dist-packages/torch2trt/torch2trt.py", line 305, in _load_from_state_dict engine_bytes = state_dict[prefix + 'engine'] KeyError: 'engine'

Also I am running this on Ubuntu Laptop machine instead of a Jetson nano. Specifications: Ubuntu 18.04 Python 3.6 TensorRT 7 (built from source) Torch 1.2 (pip install)

Also is capturing total num of frames and dividing by inference time for one frame a correct way to calculate FPS?

Regards Aditya

[JuniorKoukponou]

Hi, I have the same error. Did you find a solution, please ? Have a good day Jr

[JuniorKoukponou]

If you still search for the solution: You may convert your resnet18_baseline_att_224x224_A_epoch_249.pth intoresnet18_baseline_att_224x224_A_epoch_249_trt.pth by running the first 6 code cells of the (live_demo)[https://github.com/NVIDIA-AI-IOT/trt_pose/blob/master/tasks/human_pose/live_demo.ipynb] notebook. Thanks Jr