apply TTA during inference using sahi

[ramdhan1989]

Is there any example code to apply TTA? I only found an example for yolov5 using torch.hub.load. However, I could not find it for other frameworks (detectron, mmdetection, etc)

[fcakyon]

Hello @ramdhan1989, can you provide a reproducible code example so that we see which functions and modules are called?

[ramdhan1989]

I apologize, my previous question seems not related with this repo. I changed my question if that would be ok.

thanks

[fcakyon]

You can find TTA examples for Detectron2, MMDetection, HuggingFace and Torchvision models here: https://github.com/obss/sahi/tree/main/demo

[ramdhan1989]

I have checked the notebooks. However, I couldn't find the syntax that used to activate or deactivate TTA during inference. would you mind elaborating more about the TTA option in the code?

thanks

[fcakyon]

What do you mean by TTA?

By default, the SAHI predict function combines standard prediction and sliced prediction results.

Set no_standard_prediction: bool = True if you want to disable standard prediction

sahi/sahi/predict.py

Line 302 in 287accf

no_standard_prediction: bool = False,

Set no_sliced_prediction: bool = True if you want to disable sliced prediction

sahi/sahi/predict.py

Line 303 in 287accf

no_sliced_prediction: bool = False,

Similarly,

Set perform_standard_pred: bool = False if you want to disable standard prediction in get_sliced_prediction function

sahi/sahi/predict.py

Line 130 in 287accf

perform_standard_pred: bool = True,

[ramdhan1989]

Ok noted
thanks for your information