Can dm365 resizer deal with interlaced video?

wei han said:

The drivers doc say like this" Continuous/On the fly mode capture mode through previewer and resizer, the driver uses bottom field and scale it up vertically by 2x to get de-interlaced video. The interlace data cannot be captured in this mode."

I feel doc is referring to what is being implemented in driver. It always gives de-interlaced video (driver internally combines even/odd fields into single frame). But, you should be able to change the driver to get interlaced frames.

Can you please provide link for the doc?

wei han said:

The top field should be throwed away directly. but I have no idea how the driver do that?

The key point is that the de-interlaced video is bad in vertical resolution.

Are you observing bad video in your set-up?

wei han said:

On the other hand, I need to use resizer to scale the video sometime. So I want to know if resizer can deal with the interlaced video directly?

I feel you should be able to do that.

Nagabhushana said:

Can you please provide link for the doc?

http://focus.ti.com/lit/an/sprs566a/sprs566a.pdf, refer to p45.

In  continuous mode, that is, ISIF(CCDC) and IPIPE(previewer and resizer) is chained together, I thought it was possible to get interlaced frames. But now, I guess IPIPEIF only send one field video tho the IPIPE for interlaced video(I am now trying to validate it ). Even if my guess is not right, I donot think the IPIPE can deal with interlaced video very well, it may not  identify interlaced input or progressive input, it may always deal with the input video as  progressive and think the video comprised of odd  field and even field is a frame video. This is my analysis.

Nagabhushana said:

Are you observing bad video in your set-up?

Below is the original video and the captured video by dm365 in continuous mode.

I really have no idea if a interlaced video can be properly dealed by IPIPE, except it can use bottom field and scale it up vertically by 2x to get de-interlaced video.

wei han,

I suggest you to capture this same frame/video from CCDC/ISIF. And then compare the video with the one you mentioned as faulty. 

Note:- The constraint mentioned in doc is referring to driver implementation which captures de-interlaced video. So, applications can't get interlaced frames.

Hi Nagabhushana,

Today I have tested the method you said, there is no problem with the video captured from ISIF directly. In fact, I would be able to solve the resolution problem in my application by running the VPFE drivers in Single Shot mode but not in Continuous mode. It is just more complex for my application, because I need to use resizer to resize video and convert the colorspace type in addition. If the two functions can be realized in continuous mode, it would be perfect.

Furthermore, for dm365, if I donot want to use the simple de-interlace method above(capture run in continuous mode, that only use bottom field), which is bad for still video, is there other method to realize the de-interlace?

Thanks.

Hi，

I have just received a De-Interlace Algorithm on DM365 from TI, which uses dual iMX in IMCOP subsystem and can satisfy my requirement.

Many thanks for help!

Hi everybody,

I have the same problem for capturing interlaced video from a tvp5150 with dm365. May you please tell me which are the changes need to be applied in drivers (capture isrs, ecc)?

Thanks

Regards,

Tommaso

wei han said:

I have just received a De-Interlace Algorithm on DM365 from TI, which uses dual iMX in IMCOP subsystem and can satisfy my requirement.

John A

You can get interlaced video by changing the oper_mode parameter in the u-boot bootargs.  You can get UYVY video but cannot get 420SEMIP video unless you capture deinterlaced.  So this is an important consideration.  If you need 420 interlaced you can use the resizer to convert the UYVY to 420 after capture.  But there is a performance hit.

You can capture interlaced 420, but need to make mods to the kernel drivers to chain the resizer in interlaced mode.  I made the mods but had issues with field order.

John A

John,

There is a basic deinterlacing module available on DM36x. It works on the hardware coprocessors. Please contact your local TI sales/FAE and i can then send it to him accordingly.

Regards,

Anshuman

I also do the De-Interlace Algorithm on dm365 , But the speed is very slow with two instance of h264 encode.

I heared of that they use different coprocessor , but why so slow parallel with the h264.

Thank  you !

Hi  Anshuman,

I also do the De-Interlace Algorithm on dm365 , But the speed is very slow with two instance of h264 encode.

I heared of that they use different coprocessor , but why so slow parallel with the h264.

Thank  you !