cpi->vbp_threshold_copy = (cpi->y_dequant[q][1] << 3) > 8000
? (cpi->y_dequant[q][1] << 3)
: 8000;
+ if (cpi->rc.high_source_sad ||
+ (cpi->use_svc && cpi->svc.high_source_sad_superframe)) {
+ cpi->vbp_threshold_sad = 0;
+ cpi->vbp_threshold_copy = 0;
+ }
}
cpi->vbp_threshold_minmax = 15 + (q >> 3);
}
int pixels_wide = 64, pixels_high = 64;
int64_t thresholds[4] = { cpi->vbp_thresholds[0], cpi->vbp_thresholds[1],
cpi->vbp_thresholds[2], cpi->vbp_thresholds[3] };
+ int scene_change_detected =
+ cpi->rc.high_source_sad ||
+ (cpi->use_svc && cpi->svc.high_source_sad_superframe);
// For the variance computation under SVC mode, we treat the frame as key if
// the reference (base layer frame) is key frame (i.e., is_key_frame == 1).
}
// If source_sad is low copy the partition without computing the y_sad.
if (x->skip_low_source_sad && cpi->sf.copy_partition_flag &&
+ !scene_change_detected &&
copy_partitioning(cpi, x, xd, mi_row, mi_col, segment_id, sb_offset)) {
x->sb_use_mv_part = 1;
if (cpi->sf.svc_use_lowres_part &&
// Index for force_split: 0 for 64x64, 1-4 for 32x32 blocks,
// 5-20 for the 16x16 blocks.
- force_split[0] = 0;
+ force_split[0] = scene_change_detected;
if (!is_key_frame) {
// In the case of spatial/temporal scalable coding, the assumption here is