[Ffmpeg-devel] Re: tracing GOP start code

Micro Mysore micromysore
Thu Jul 28 02:11:40 CEST 2005

It was a good start. I finally found the code !! I was trying to find
the code which resets the GOP upon detecting a scene change. I came
across a paper published for MPEG in 1996 called SDMAD (Signed
difference of mean absolute difference). Looks like the equations used
here are way diff.

in libavcodec/motion_est.c

01179             if (vard <= 64 || vard < varc) { //FIXME
01180                 c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
01181             }else{
01182                 c->scene_change_score+= s->qscale;
01183             }

and further used in 

05301     if(s->me.scene_change_score >
s->avctx->scenechange_threshold && s->pict_type == P_TYPE){
05302         s->pict_type= I_TYPE;
05303         for(i=0; i<s->mb_stride*s->mb_height; i++)
05304             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
05305 //printf("Scene change detected, encoding as I Frame %d %d\n",
s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
05306     }

Now, what is done is clear. Is there some document which describes, why ? 



On Wednesday 27 July 2005 04:02, Micro Mysore wrote:
> Hello,
> I want to find the method used to determine the start of GOP ..

the code surrounding `grep -i gop mpegvideo.c` should be a start

> I am trying to find if ffmpeg uses constant length of GOP or if not,
> what algorithm it uses to find the scene change.

that depends upon various parameters


More information about the ffmpeg-devel mailing list