When doing yolox/trt_evaluate.sh for fp32 tensorrt model, what's the reason this error occurred? how can i fix it?

[NutshellLee]

trt_model checkpoints/tensorrt/yolox_x_8x8_300e_coco_20211126_140254-1ef88d67.trt loading annotations into memory... Done (t=0.55s) creating index... index created! [ ] 0/4952, elapsed: 0s, ETA:Traceback (most recent call last): File "tools/2d/evaluate_trt.py", line 142, in main() File "tools/2d/evaluate_trt.py", line 103, in main engine, context, input_shapes=input_shapes_, output_shapes=output_shapes_ File "./det2trt/utils/tensorrt.py", line 49, in allocate_buffers context.set_binding_shape(binding_id, dims) AttributeError: 'NoneType' object has no attribute 'set_binding_shape'

[DerryHub]

You can set https://github.com/DerryHub/BEVFormer_tensorrt/blob/1b39aac2fc260b65347a9479bb4e85905c37dff2/tools/2d/evaluate_trt.py#L40 to trt.Logger.VERBOSE to see how the error happened.

[NutshellLee]

Hi, I believe it's a bug. When evaluating fp32 trt yolox-x model, I found that the context is none, which means engine.create_execution_context() returns none. But the engine is not none. It doesn't happen in other precision types of trt, only in fp32 trt. fp32 pth model works fine too.