'std::bad_alloc' Error on Changing Image Dimensions in 'config.yaml'

[parsahemmasi]

Description
When I attempt to change the image_width and image_height values in the config.yaml file from the default 320x240 to any other dimensions, the engine file builds successfully. However, during inference, the following error occurs:

terminate called after throwing an instance of 'std::bad_alloc'
what(): std::bad_alloc
Aborted(core dumped)

Steps to Reproduce

1. Clone the repository.
2. Modify the config.yaml file to change image_width and image_height values to something other than 320x240 (e.g., 1280x960).
3. Run the superpointglue_image with two 1280x960 images as input arguments.

Expected Behavior
The inference should run successfully with the updated image dimensions.

Actual Behavior
The process terminates with a std::bad_alloc error, indicating a memory allocation issue.

Environment

• OS: Ubuntu 20.04.6 LTS
• Device: NVIDIA Tegra Orin (nvgpu)/integrated
• CUDA Version: 11.4.315
• CUDNN Version: 8.6.0.166
• TensorRT Version: 8.5.2.2

Additional Context

• The default 320x240 dimensions work without any issues.
• It appears that the memory allocation fails when dimensions are altered, which might indicate a problem with the engine's handling of different input sizes.

[vickersmt]

I think you should modify the size of the dynamic input in superpoint.cpp and superglue.cpp，search"profile->setDimensions"

[parsahemmasi]

I think you should modify the size of the dynamic input in superpoint.cpp and superglue.cpp，search"profile->setDimensions"

I have tried but this time I got an error about building engine.
Would you please explain more?
And what is your recommended modification for inferencing two 1280x960 images?