Questions about the transformation from (H/4, W/4) feature coordinate to the (H, W) pixel coordinate.

[Lightingooo]

Thank you for your excellent work. During my reading, I found you used 4 * (coords + 0.5) to transform the (H/4, W/4) feature coordinate to the (H, W) pixel coordinate in here. It seems that 4 * coords + 1.5 could be more suitable if we define both the coordinate in the following form: (x, y) represents the center for the corresponding pixel block, which means for both coordinate the low range is -0.5. Could you please give some explainations on the coordinate defination?

Moreover, here and the following lines transform the (H/4, W/4) feature coordinate of frame i's patch into frame j. It directionly applys to the (H/4, W/4) feature coordinate instead of (H, W) coordinate, which is right under the assumption that the intrinsics correspondingly scale to 1/4. However, the intrinsics transformation also should consider the 0.5 pixel shift.