I noticed if we use the mmsegmentation demo code to predict on image using for example psnet is different from if we just write a standard code to load the model and predict as the output size is downsized by 8 and resize function doesn’t simulate the output from mmsegmentation.
I do understand there is a preprocessing and post processing steps in mmsegmentation inference.
can we obtain similar result using libtorch for example?
Hi all,
I noticed if we use the mmsegmentation demo code to predict on image using for example psnet is different from if we just write a standard code to load the model and predict as the output size is downsized by 8 and resize function doesn’t simulate the output from mmsegmentation.
I do understand there is a preprocessing and post processing steps in mmsegmentation inference.
can we obtain similar result using libtorch for example?