Paper corrections

Fig. 7 (The minimum block SAD computation of the HW ABMOF)

This figure in paper is not quite correct because there was a bug in the latex code that generated it.

There is a mistake in the figure. We used latex Tikz to draw this figure, so there might be some bugs in that code.

The idea for calculating the column SAD in PE5 RowSummer is in order as shown in this figure which means:

1. At clock 1: outer column 1 and inner column 1 -> output column 1

2. At clock 2: outer column 2 and inner column 2 -> output column 2

3. …

4. At clock 9: outer column 9 and inner column 9 -> output column 9

This figure also showed that ABMOF uses signed accumulated events in the paper. Our experiments showed that signed polarity did not significantly improve flow accuracy, so we used the extra bit to accumulate more signed events. Therefore the event counts in this figure should be unsigned ints.

We thank Biyan Zhou and Prof. Arindam Basu of NTU Singapore for pointing out this mistake.

"In Fig 7 (The minimum block SAD computation of the HW ABMOF), I found the 1st column of 'Column SAD FIFO' is calculated from the 1st column of 'Outer block column FIFO' and the 1st column of ' Inner block column FIFO' . But other column of 'Column SAD FIFO' is not calcuated by corresponding column of 'Outer block column FIFO' and ' Inner block column FIFO' respectively. Can I assume there is an error with calculated values shown in Fig. 7? The values in question can be seen below (extracted from Fig. 7). If not, could you help me to understand how does it work?