Get Binary Mask as Inference Output

[d3tk]

Hello, I am using the torch inference and the inference outputs the heatmap superimposed on the original image and then a segmentation. Is there anyway to get just the binary mask as the output?

[blaz-r]

Hello. I don't quite understand your question. You can get binary mask from object returned by inferencer. It's a property called pred_mask.

[d3tk]

I am using anomalib\tools\inference\torch_inference.py from the command line. I set the task to segmentation. Is there another way you'd recommend doing inference?

[blaz-r]

You will need to modify the code then, as I believe that default there is only simple visualization (the one where heat map is superimposed) and full (where you get the grid).
You will need to add output = predictions.pred_mask and avoid using visualizer.

[d3tk]

Thank you that fixed it. changed line 84 in torch_inferencer.py to be ouput=predictions.pred_mask then I just save it with cv2