About sliding_window_inference

[uto-lt]

Hi, @a-parida12
Could you please show your sample code about using this function in multiple inputs and single output scenarios

[a-parida12]

My model forward method required 2 inputs. I stacked my 2 inputs in the channel dimension and passed it thru a function called bunched_infer which called the actual forward call. See the example below-

def bunched_infer(inp):
        out = model.forward(inp[:,inp.shape[1]//2:,...], inp[:, :inp.shape[1]//2,...])
        return out
 
 inp1 = data1['image']
 inp2 = data2['image']
 stack_inp = torch.cat([inp1, inp2], axis = 1) 
 
 pred1= sliding_window_inference(stack_inp, roi_size=(16, 16, 16), sw_batch_size=1, predictor=bunched_infer, overlap=0.50, mode='gaussian')     

This answer works with the original query I had #4318 (reply in thread)

[uto-lt]

Nice code! Thanks for your reply.