can't test forward

[ahuxjz]

prediction[:,:,:2]+=x_y_offset
RuntimeError: The expanded size of the tensor (507) must match the existing size (13520) at non-singleton dimension 1

[harpalsahota]

Got a similar error, but mine is happens in the predict_transform method, when calling prediction.view

RuntimeError: shape '[1, 255, 3025]' is invalid for input of size 689520

Printing the inputs to prediction.view I get:

batch_size, bbox_attrs, num_anchors, grid_size
1 85 3 13
1 85 3 26
1 85 3 55 <--- crashes here

[harpalsahota]

I think the issue is in the cfg file. The original height and width was 416 but now is 608. So

stride = in_dim // prediction.size(2)

went from 416 // 52 = 8 to 608 // 52 = 11. Therefore:

grid_size = in_dim // stride

went from 416 // 8 = 52 to 608 // 11 = 55. The latter has the incorrect grid size this causing the shape error here:

prediction = prediction.view(batch_size, bbox_attrs * num_anchors, grid_size * grid_size)

Would the correct thing to do be changing the height and width back to 416?

[AbanoubMamdouh]

@harpalsahota The most important thing is that the input image dimensions which you resize in the function:

get_test_input()

matches with the dimensions of the image in the cfg file. I've tried it myself and it worked fine.