ffmpeg,X264编码结果I帧QP比P帧还大

enc_ctx->profile =FF_PROFILE_H264_MAIN ;
enc_ctx->time_base.den = 24;
enc_ctx->time_base.num = 1;
enc_ctx->gop_size = 8; /* emit one intra frame every twelve frames at most */
enc_ctx->pix_fmt = AV_PIX_FMT_YUV420P;
enc_ctx->max_b_frames = 0;
enc_ctx->has_b_frames = 0;
av_opt_set(enc_ctx->priv_data, "preset", "slow", 0); ///
av_opt_set(enc_ctx->priv_data, "tune", "zerolatency", 0);

enc_ctx->bit_rate = br ;
enc_ctx->rc_min_rate = br ;
enc_ctx->rc_max_rate = br ;
enc_ctx->bit_rate_tolerance = br;
enc_ctx->rc_buffer_size = br;
enc_ctx->rc_initial_buffer_occupancy = enc_ctx->rc_buffer_size*3/4;
enc_ctx->rc_buffer_aggressivity = (float)1.0;
enc_ctx->rc_initial_cplx = 0.5;
enc_ctx->i_quant_factor = -1;

基本设置，固定码率150k，enc_ctx->i_quant_factor默认为 -1，编码结果

[libx264 @ 005cbb40] frame I:38 Avg QP:36.34 size: 4044
[libx264 @ 005cbb40] frame P:262 Avg QP:35.27 size: 339

I的质量比P还差。

ffmpeg中i_quant_factor的解释

/**
* qscale factor between P and I-frames
* If > 0 then the last p frame quantizer will be used (q= lastp_q*factor+offset).
* If < 0 then normal ratecontrol will be done (q= -normal_q*factor+offset).
* - encoding: Set by user.
* - decoding: unused
*/
float i_quant_factor;

enc_ctx->i_quant_factor = -0.7;时：

[libx264 @ 006ecb40] frame I:38 Avg QP:36.11 size: 4067
[libx264 @ 006ecb40] frame P:262 Avg QP:34.89 size: 337

enc_ctx->i_quant_factor = -1.4;时

[libx264 @ 0276cb40] frame I:38 Avg QP:35.34 size: 4439
[libx264 @ 0276cb40] frame P:262 Avg QP:35.17 size: 280

enc_ctx->i_quant_factor = -2;时

[libx264 @ 02beda40] frame I:38 Avg QP:34.59 size: 4155
[libx264 @ 02beda40] frame P:262 Avg QP:33.71 size: 323

i_quant_factor 相当于x264的参数ipratio

默认值：1.40

修改I帧量化值相比P帧量化值的目标平均增量。越大的值会提高I帧的品质。

原因估计：应该是i_quant_factor 绝对值越大，I帧质量越好。还有猜测可能是码率不够，限制了I帧的比特数。

设置固定码率300k时：

[libx264 @ 016dbb40] frame I:38 Avg QP:29.21 size: 6367
[libx264 @ 016dbb40] frame P:262 Avg QP:29.60 size: 622