动漫国调deadzone应该是Myth大大的习惯而已:) trellis=2下deadzone无用
另外小弟不太理解Psy trellis :)
Psy trellis: bias in favor of higher AC coefficients in the reconstructed frame.
- if( h->mb.i_psy_trellis && i && !dc && ctx_block_cat != DCT_CHROMA_AC )
- {
- int orig_coef = (i_coefs == 64) ? h->mb.pic.fenc_dct8[idx][zigzag[i]] : h->mb.pic.fenc_dct4[idx][zigzag[i]];
- int predicted_coef = orig_coef - i_coef * signs[i];
- int psy_value = h->mb.i_psy_trellis * abs(predicted_coef + unquant_abs_level * signs[i]);
- int psy_weight = (i_coefs == 64) ? x264_dct8_weight_tab[zigzag[i]] : x264_dct4_weight_tab[zigzag[i]];
- ssd = (int64_t)d*d * coef_weight[i] - psy_weight * psy_value;
- }
- else
- ssd = (int64_t)d*d * (dc?256:coef_weight[i]);
完全看不懂[/ku]
似乎最终影响了ssd值
/* Psy RD distortion metric: SSD plus "Absolute Difference of Complexities" */
/* SATD and SA8D are used to measure block complexity. */
/* The difference between SATD and SA8D scores are both used to avoid bias from the DCT size. Using SATD */
/* only, for example, results in overusage of 8x8dct, while the opposite occurs when using SA8D. *
