UG DM35x spruf71a Figure 74 Register Settings for Frame Division Mode - V

Hi Williem,

I am putting down the values as below:

h1 = h2 = 2368/2 = 1184

H1 = H2 = 2560/2 = 1280

o = 20 as resize ratio is greater than 1/10

d = 2368/2560*256 = 236

For right side vertical slice, calculation are as below:

xo = h1-o = 1184 - 20 = 1164

X0 = 1260 (calculations based on formula)

x1 = 1164 (based on calculation)

input_h_position = 0

output_h_position = 20

output_h_size = 1280

h_initial_phase = 384

You need to change the input address and output address for the right block of the resizer. The address would be calculated as :

Input Addr of right slice = Input Addr of Left Slice + 1164 pixels

Output Addr of right slice = Output Addr of Left Slice + 1280 pixels

Hope this helps.

Regards,

Anshuman

Hi Anshuman,

Thank you very much for the reply.

I am still having difficulty with the calculation of X0.

The formula in spruf71a:

X0 = 2 * ceil((x0 * 256) / (d / 2))

= 2 * ceil((1164 * 256) / (236 / 2))

= 2 * ceil(297984 / 118)

= 2 * 2526

= 5052

Should the correct formula be:

X0 = 2 * ceil(((x0 * 256) / d) / 2)

= 2 * ceil(((1164 * 256) / 236) / 2)

= 2 * ceil(1262.6 / 2)

= 2 * 632

= 1264

Then I get:

input_h_pos = 0

output_h_pos = 16

h_init_phase = 320

How does the resizer know that it is to use Frame Division Mode V? Is RSZ_H_PHS the only parameter that specifies that it should be using Frame Division Mode V?

Thanks

Willem

Hi Willem,

Willem Swart said:

Should the correct formula be:

X0 = 2 * ceil(((x0 * 256) / d) / 2)

Yes, this formula is correct. I somehow used the same formula when computing my values and did not realize the difference till you pointed out. I will feed it back to our documentation team to see if they can fix it. BTW, i used value of "d" as 236.8 instead of 236, so our numbers would be looking a little different. But as you have done it, using 236 is correct.

Willem Swart said:

How does the resizer know that it is to use Frame Division Mode V? Is RSZ_H_PHS the only parameter that specifies that it should be using Frame Division Mode V?

 

 

Yes, RSZ_H_PHS tells that there is a Frame Divison - V mode being used.

Please mark the answer as verified if it worked for you, so that others can refer to it.

Regards,

Anshuman

Hi Anshuman,

Thanks for the help, I haven't got it working yet, but am making progress, I would like to understand how it is working so that I can fix what I am doing wrong.

Q1. In general, how does the resizer know how many pixels horizontally to use from the input as there is no RSZ_I_HSZ parameter?

Q2. What is RSZ_H_PHS? How is it applied by the resizer? I am not sure of the meaning of RSZ_H_PHS. Elsewhere I have seen it described as the "mixing ratio of the first input line". Is this the ratio of the overlap in the input vs the overlap in the output?

Q3: For some of the resolutions that I have to resize, e.g. 1280x960 -> 2560x1944, the calculated RSZ_H_PHS is 0. Does this mean Frame Division Mode V will not be used if the parameter RSZ_H_PHS is 0?

Q4:

Processing the left hand side:

RSZ_O_HSZ = 1284 - 1, which is greater than H1, so I assume it is writing some overlap (about 4 pixels) to the output which the right hand side processing will take into account to prevent seaming in the middle of the image using the RSZ_H_PHS value?

Processing the right hand side:

The input buffer pointer is adjusted to be the start of the overlap position so it is going to process (h2 + overlap) pixels I assume and write that to the output image.

The output buffer pointer is adjusted to be 1280 (effectively start position of H2), then RSZ_O_HST  is a 16 offset from 1280?. So in the output it will start processing at position 1280 + 16? (But RSZ_O_HSZ = 1280, and if we start at 1280 + 16 the RSZ_O_HSZ should be less than 1280?).

Does the resizer use the RSZ_H_PHS parameter to run a filter over the overlapped pixels?

How does the RSZ_H_PHS = 320 come into it?

I would have thought processing the right hand side, RSZ_O_HSZ should be X0 size? (the input h2+overlap scaled to the output H2 +overlap?)

So as you can see I really don't have a good idea of what is going on and would appreciate it if you clear things up to me as I would like to get this working.

Thanks

Willem

Hi,

I'm still having a problem with the Frame division mode-V, the vertical seam looks ok, but the right-hand side strip's right edge of the image is smeared/blurred as per the image below. Any ideas as what could be wrong or could be causing it would be appreciated.

Source image: 2368x1776

Target image: 2560x1944

h = 2368; h1 = h2 = 1184

H = 2560; H1 = H2 = 1280

overlap = 20

d = 2368/2560 * 256 = 236

x0 = 2368 - (1184 + 20) = 1164

X0 = 2 * ceil(((x0 * 256)/d)/2) = 1264

x1 = 1164

Left side:

I_HST = 0

O_HST = 0

O_HSZ = 1284

H_PHS = 0

Right side:

I_HSZ = 1204

I_HST = 0

O_HST = 16

O_HSZ = 1280

H_PHS = 320

Advance input buffer pointer by 1164

Advance output buffer pointer by 1280

The smeared/blurred strip for above example is approximately 16 pixels but I have no idea what causes it. I suspect it has something to do with the overlap size or filter.

Image:

Thanks

Willem