TDA4VM: Queries Regarding H.264 Encoder

Hi Dwarakesh,

Dwarakesh Radhakrishnan said:

we use V4L2_CID_MPEG_VIDEO_GOP_SIZE for that, is it correct?

Gop Size is properly tested and you can use that.

Dwarakesh Radhakrishnan said:

we are using V4L2_CID_MPEG_VIDEO_H264_I_PERIOD, but is it valid?

I_PERIOD is not tested properly on our side.

Dwarakesh Radhakrishnan said:

-- base profile

If you are trying to force the driver to base profile, You can use the below patch. This patch is a part of SDK 8.5, which will be released in the upcoming month.

1 file changed, 2 insertions(+), 2 deletions(-)
drivers/media/platform/vxe-vxd/encoder/vxe_enc.c | 4 ++--

modified   drivers/media/platform/vxe-vxd/encoder/vxe_enc.c
@@ -353,10 +353,10 @@ void vxe_fill_default_params(struct vxe_enc_ctx *ctx)
  ctx->vparams.slices_per_picture = 1;
  ctx->vparams.is_interleaved = FALSE;
  ctx->vparams.constrained_intra = FALSE;
- ctx->vparams.h264_8x8 = TRUE;
+ ctx->vparams.h264_8x8 = FALSE;
  ctx->vparams.bottom_field_first = FALSE;
  ctx->vparams.arbitrary_so = FALSE;
- ctx->vparams.cabac_enabled = TRUE;
+ ctx->vparams.cabac_enabled = FALSE;
  ctx->vparams.cabac_bin_limit = 2800;
  ctx->vparams.cabac_bin_flex = 2800;
  ctx->vparams.deblock_idc = 0;

Dwarakesh Radhakrishnan said:

   unsigned short intra_refresh;

I have checked this, and I dont think there is any support in the driver for this.

Dwarakesh Radhakrishnan said:

3. In driver I see that NO_CR Topaz fw is used. Can we use CBR for constant bitrate? What should we do for that?

    ret = topaz_upload_firmware(ctx, IMG_CODEC_H264_NO_RC);

I think, I will need some more time to get back to you on this.

Regards

Nikshith

Hi Dwarakesh,

Nikshith Reddy Kolan said:

Can we use CBR for constant bitrate?

Yes, it is supported. 

Nikshith Reddy Kolan said:

What should we do for that?

The value can be passed through the Gstreamer pipeline.

regards

Nikshith

Hi Nikshith,

Here two sdp. we want TI encoder to produce same 7875.reference_iccd.txt5148.ti_iccd.txt SPS described in Reference sdp. Currently we have settings presented in ti_iccd.sdp.

Parsed in readable format fields are applied in txt file.

Settings same as provided before in ticket

Hi Dwarakesh,

Regarding the questions mailed,

3) We observe that when we set the bitrate of the stream to 1M or similar we still get an actual bitrate of 2.5M. If bitrate value is set 2.5M then actual is more as well. On another platform we used actual bitrate as requested. I supposed it is due to "no_rc" (no rate control) firmware used in Topaz but I might be wrong.

On TDA4VM SDK 8.4 Linux, bitrate is working as expected using standalone gstreamer command.

Can you share more details on how you are encoding the file?

Are you using a camera input or filesrc?

If possible can you share the raw files as well?

We are working on other 2 questions as well with priority. Will update the details soon.

Regards

Nikshith

Hi Dwarakesh, 

Dwarakesh Radhakrishnan said:

1. We want to use h264 encoding with following settings:

- intra refresh period is not supported and using V4L2_CID_MPEG_VIDEO_H264_I_PERIOD for intra refresh is not valid

- In TDA4VM driver, V4L2_CID_MPEG_VIDEO_GOP_SIZE is used to set IDR_PERIOD.

- Rate control mode settings are not supported and VBR is enabled by default.

- In order to use base profile, please use the patch shared in the above comment. High profile is enabled by default in the driver. 

Dwarakesh Radhakrishnan said:

2. In Topaz API header there is field:

    unsigned short intra_refresh;

- The variable is just defined but has no implementation in the driver.

Regards

Nikshith

Hi Dwarakesh,

1. Can you share more details on how you are creating the sdp file using TI encoder like gstreamer command or steps to reproduce the issue

2. There are many parameters which are different. Are there any parameters with priority?

Regards

Nikshith

Hi Nikshith Reddy Kolan,

I have below queries and need your help to answer these -

1 - Is CBR rate control supported ??

2 - Any link page or user guide to know all the supported Encoder features ?? E,g Link

3 - Setting the below parameters, Rate control internally takes cares the parameters based on the input stream ??

    ctx->rc.initial_qp_i = 0;

    ctx->rc.initial_qp_p = 0;

    ctx->rc.initial_qp_b = 0;

    ctx->rc.min_qp = 0;

    ctx->rc.max_qp = 0;

    ctx->rc.rc_enable = TRUE;

Regards

Gajanan

Hi Gajanan,

We have raised a request with our partner team. Will update you once we get the response.

Regards

Nikshith

Hi Nikshith,

Thanks for raising this query with partner team. I have another following question on the encoding parameters -

1 - If I change the below parameters, the generated stream is non-compliance and not able to decode properly

2 - If I change the below parameters, the generated stream is non-compliance and not able to decode properly

Can you let me know what changes to be done to get the above parameters reflect and generate decodable streams ???

Regards

Gajanan

Hi Nikshith,

When I enabled the CBR mode, we observed the bitrate achieved is not consistent and under achieved for the same video contents.

Example, if we set 2.5Mbps, we are getting 1.5Mbps. The bitrate graph is shown below -




Regards

Gajanan

Hi Gajanan,

Gajanan Maske said:

2 - Any link page or user guide to know all the supported Encoder features ??

All the supported encoder features are mentioned in the documentation. please use the below link.

https://software-dl.ti.com/jacinto7/esd/processor-sdk-linux-jacinto7/08_04_00_11/exports/docs/linux/Foundational_Components_Multimedia_D5520_VXE384.html#v4l2-video-encoder

Gajanan Maske said:

When I enabled the CBR mode, we observed the bitrate achieved is not consistent and under achieved for the same video contents.

We are already testing by changing the rc_mode to CBR and facing similar kind of inconsistencies as you are facing. We will get back to you.

Regards

Nikshith

Hi Gajanan,

Can you give more clarity on the graph attached.

Gajanan Maske said:

Example, if we set 2.5Mbps, we are getting 1.5Mbps. The bitrate graph is shown below -

1. What is X-axis and Y-axis denoting?

2. What is the tool used to generate the graph. Is it open-source?

3. You have mentioned a bitrate of 1.5Mbps. Can you let us know, how did you infer that from the graph attached?

4. The graph seems to have periodic nature. Can you share the raw .yuv file?

HI Nikshith,

Nikshith Reddy Kolan said:

1. What is X-axis and Y-axis denoting?

X-axis ==> Frame Number

Y-axis ==> Bytes consumed

Nikshith Reddy Kolan said:

2. What is the tool used to generate the graph. Is it open-source?

github.com/.../releases

Nikshith Reddy Kolan said:

3. You have mentioned a bitrate of 1.5Mbps. Can you let us know, how did you infer that from the graph attached?

If you open the genrated bit-stream in above shared tool, you will get these details

Nikshith Reddy Kolan said:

4. The graph seems to have periodic nature. Can you share the raw .yuv file?

This behaviour is observed with real time capture and encoding. So I dont have yuv available. And also encoded stream I cannot share it here due to confidentiality.  Since you are able to reproduce the issue at your end, I don't see any requirement of sharing this.

Kindly let me know if you need more details.

Regards

Gajanan

Hi Gajanan,

The raw file is taken from this location. https://github.com/tianchenghe2017/VTM/blob/master/BasketballPass_416x240_50.yuv

the above file is encoded both using VBR and CBR. When opened using the above tool you have shared, the graphs are different for VBR and CBR and CBR is working fine.CBR doesn't mean that every frame will be of same bitrate but, every set of frames grouped together have same mean bitrate. 

Gajanan Maske said:

if we set 2.5Mbps, we are getting 1.5Mbps. The bitrate graph is shown below -

But still could not understand how you infered 1.5 mbps bitrate from the tool. please provide more details on how to see bitrate in the tool. When checked using ffprobe, we can see the bitrate.

Regards

Nikshith

Hi Nikshith,

Nikshith Reddy Kolan said:

the above file is encoded both using VBR and CBR. When opened using the above tool you have shared, the graphs are different for VBR and CBR and CBR is working fine.CBR doesn't mean that every frame will be of same bitrate but, every set of frames grouped together have same mean bitrate. 

 ==> I will share you more details to on this

Nikshith Reddy Kolan said:

But still could not understand how you infered 1.5 mbps bitrate from the tool. please provide more details on how to see bitrate in the tool. When checked using ffprobe, we can see the bitrate.

==> You can use H264 Analyser to see the bitrate for the generate stream

Hi Nikshith,

Please use the below attached files -



8865.txt


2364.txt

Rename the above RAW video files from .txt to .yuv (e.g 8865.txt to 1.yuv & 2364.txt to 2.yuv)


Create the input raw file combination of black and actual raw video -

cat 1.yuv 2.yuv 1.yuv 2.yuv 1.yuv 2.yuv 1.yuv 2.yuv 1.yuv 2.yuv 1.yuv 2.yuv 1.yuv 2.yuv 1.yuv 2.yuv > input_704x512_nv12.yuv


Execute the below to generate CBR and VBR streams (Note the driver needs to be changed to change the modes CBR/VBR).
BITRATE is 2.5mbps, gop is 30, intra interval is 5, nv12 format

./tienc_encode -i input_704x512_nv12.yuv -w 704 -h 512 -b 2500000 -p 5 -g 30 -o output_2.5Mbps_cbr.264

./tienc_encode -i input_704x512_nv12.yuv -w 704 -h 512 -b 2500000 -p 5 -g 30 -o output_2.5Mbps_vbr.264

The result summary is below -

For CBR the encoded files bit-rate is under achieved and VBR bit-rate is overshoot

 CBR - 1.56Mbps (Underachieved)

VBR - 5.3Mbps (Overshoot)

Regards

Gajanan

Hi Gajanan,

Nikshith Reddy Kolan said:

The raw file is taken from this location. https://github.com/tianchenghe2017/VTM/blob/master/BasketballPass_416x240_50.yuv

CBR and VBR are working fine with the above basket ball video.

Gajanan Maske said:

For CBR the encoded files bit-rate is under achieved and VBR bit-rate is overshoot

 CBR - 1.56Mbps (Underachieved)

VBR - 5.3Mbps (Overshoot)

We could replicate the issue at our end. Saw under achievement in CBR and Overshoot in VBR. We are debugging the issue.

Regards

Nikshith

Hi Nikshith,

Also please reply to this query as well - 

Gajanan Maske said:

I have another following question on the encoding parameters -

1 - If I change the below parameters, the generated stream is non-compliance and not able to decode properly

ctx->sh_params.log2_max_pic_order_cnt = 8; //default value is 6

2 - If I change the below parameters, the generated stream is non-compliance and not able to decode properly

generate_ue(mtx_header, element_pointers, 4); /* log2_max_frame_num_minus4 = 1, default value is 1

Can you let me know what changes to be done to get the above parameters reflect and generate decodable streams ???

Regards

Gajanan