omron-sinicx / neural-astar

Hi All,

I have attempted to use the repo to perform neural A* search on a rectangle like graph with size (8, 1, 32, 64). However it seems that 'get_heuristic' will output an einsum error. I am wondering if there's any potential fix to this problem

Hi, thanks for your comment!

In fact the current implementation assumes input images to be square.

But extending it to non-square maps wouldn’t be that hard. I believe that specifying size1, size2 = goal_maps.shape[-2], goal_maps.shape[-1] and fixing the following two lines will solve the problem for the get_heuristic function.

neural-astar/src/neural_astar/planner/differentiable_astar.py

Lines 39 to 40 in 719b579

    
           num_samples, size = goal_maps.shape[0], goal_maps.shape[-1] 
        
           grid = torch.meshgrid(torch.arange(0, size), torch.arange(0, size))

I will fix it in the next update. Thank you!

Now supported by #18

	num_samples, size = goal_maps.shape[0], goal_maps.shape[-1]
	grid = torch.meshgrid(torch.arange(0, size), torch.arange(0, size))

Possibility of non-square map