I propose adding functionality to Gigapixel AI to support the import and utilization of user-provided ONNX models. An optional conversion to TensorRT format would further enhance this feature. This would facilitate direct comparisons between Gigapixel AI’s proprietary models and open-source/user-trained alternatives within the same application environment, leveraging Gigapixel AI’s existing features such as tiling and face recovery. As Gigapixel AI already utilizes ONNX and TensorRT models, the implementation of this feature could be relatively straightforward. This enhancement would provide significant value to users interested in exploring and experimenting with different models.
