|
|
| KlsHa |
| Posted: Apr 11 2004, 09:24 PM |
 |
|
Unregistered
|
dear team,
after reading in this forum, I got the kick to retry AVS tools.
Up to now, I use(d) VD, smart DI and smart SM. I have know tested with KERNEL_DI and PEACH and have to say, that the throughput is tremendous: 7 fps improved to somewhat about 28 fps. These results really force one to dig deeper.
The snag are now the AVS decombing filters; maybe I am wrong, but each of the tested tools has in sum a drawback against smart DI (VD). I tested Decome (Telecine resp. FieldDeinterlace) and Kernel DI. Here are my findings:
This is the link to the frame shots and the 2 avi's:
http://people.freenet.de/KLAUSG-HANNOVER/a...avs_downld.html
- DI-Test (Pic15) and Colo30 (Pic07); the jpg's are however taken from Mainconcept Codec and full quality
legend:
smDI=smart Deint (VD); TC=Telecide; FD = Fielddeinterlace; KERN=kernel deint.
a) .. d) ranking according the subject
parameters: TC: Ord=0; Guide=2; Post=2; vth=45; dth=13
FD: Ord=0; thr=5; dth=17
Case A:
The sequence 1) .. 5) refers to "DI-Test-PIC15.avi" (3.5 GB if needed, compression level 15)
1) frame 12:
KERN is not bad, but has highest grain in this shot.
TC shows slight colored specks; this effect already reduced with dth=13; with default settings specks with more color
2) frame 18 has a defect line within one or both fields.
Telecide and FD do badly cope with this problem.
KERN is perfect and smart DI not bad.
3) frame 21:
FD and TC produce strong wawy (meander) edges at upper door edge;
smDI is very good and KERN not bad.
This issue is certainly important, because a lot of footage is with diagonal (<> 0 and 90 degrees) lines.
4) frame 26:
SmDI and FD render a smoth door blade;
TC and KERN show some artefarcts and tend to color stains;
5) frame 37:
SmDI and FD render a smoth door blade;
TC and KERN again show artefarcts and tend to color stains;
FD produces rather good vertical door edges and lines (all together)
KERN projects a dark vertical shadow on the left inner door blade edge.
Case B
Another subject is either my bad understanding of TeleCide or we have another problem there:
(taken from Colo-07PIC.avi, 1.6GB - higher compression)
This footage requires post=0 setting to force progressive processing.
Allowing postprocessing (POST = 2) results in a deinterlaced frames, which produce with this footage an ugly jumping of the doublelines at the train body.
As I sometimes experience progressive footage with short interlaced intersections, I hoped for a kind of foolprove automatic with Telecide. But as I can see here: Telecide (post=0) would not help.
In my link you find the avi videos separately, the jpg's zipped according case A and case B and the avs files zipped - due to 1,5 MB upload limit, the case A units are in 2 rar files each :-(
Thanks a lot for your comments and guidance; as welI as for further enhancements in future.
Have a happy easter!
/Klaus
|
|
| neuron2 |
| Posted: May 12 2004, 05:05 AM |
|
|
Advanced Member
Group: Members
Posts: 1244
Member No.: 5294
Joined: 18-July 03
|
Your application of Telecide (IVTC) to pure interlaced video shows that you do not know what you are doing.
I suggest that you learn the fundamentals before conducting shootouts.
BTW, there is an Avisynth port of SmartDeinterlacer.
Hope I am not too frank in my comments. |
|
|
