Using Monai Sliding Window with a YOLOv8 Segmentation model

[Mehreganbeiki]

Hi,

So I have been trying to use the Monai sliding window on my test images. Here in the code, I load my Yolov8-seg model, which I trained earlier, and also read an image, convert it to a torch tensor, unsqueeze it to have the specific shape needed, and pass it to the sliding window. I am getting the following error, which I assume is because something with the model is not compatible with what function requires. I am working with RGB 2D images.

roi_size = 1312
sw_batch_size = 8
overlap = 0.3
model = YOLO("best.pt")
test_file_path = "L 20.jpg"

image = cv2.imread(test_file_path)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
transform = transforms.ToTensor()
tensor_image = transform(image)
tensor_image = tensor_image.unsqueeze(0)
sliding_window = sliding_window_inference(tensor_image, roi_size, 1, model)

[KumoLiu]

Hi @Mehreganbeiki, the main issue is that type of the prediction from the model is incompatible with the sliding_window_inference, you can simply convert it to a tensor in the model to fix the issue, thanks.