summaryrefslogtreecommitdiff
path: root/vpx
diff options
context:
space:
mode:
Diffstat (limited to 'vpx')
-rw-r--r--vpx/internal/vpx_psnr.h34
-rw-r--r--vpx/src/svc_encodeframe.c156
-rw-r--r--vpx/src/vpx_encoder.c42
-rw-r--r--vpx/src/vpx_psnr.c24
-rw-r--r--vpx/svc_context.h3
-rw-r--r--vpx/vpx_codec.mk2
-rw-r--r--vpx/vpx_encoder.h7
7 files changed, 225 insertions, 43 deletions
diff --git a/vpx/internal/vpx_psnr.h b/vpx/internal/vpx_psnr.h
new file mode 100644
index 000000000..07d81bb8d
--- /dev/null
+++ b/vpx/internal/vpx_psnr.h
@@ -0,0 +1,34 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VPX_INTERNAL_VPX_PSNR_H_
+#define VPX_INTERNAL_VPX_PSNR_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// TODO(dkovalev) change vpx_sse_to_psnr signature: double -> int64_t
+
+/*!\brief Converts SSE to PSNR
+ *
+ * Converts sum of squared errros (SSE) to peak signal-to-noise ratio (PNSR).
+ *
+ * \param[in] samples Number of samples
+ * \param[in] peak Max sample value
+ * \param[in] sse Sum of squared errors
+ */
+double vpx_sse_to_psnr(double samples, double peak, double sse);
+
+#ifdef __cplusplus
+} // extern "C"
+#endif
+
+#endif // VPX_INTERNAL_VPX_PSNR_H_
diff --git a/vpx/src/svc_encodeframe.c b/vpx/src/svc_encodeframe.c
index adce47637..c7837244f 100644
--- a/vpx/src/svc_encodeframe.c
+++ b/vpx/src/svc_encodeframe.c
@@ -13,6 +13,7 @@
* VP9 SVC encoding support via libvpx
*/
+#include <math.h>
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>
@@ -40,6 +41,7 @@ _CRTIMP char *__cdecl strtok_s(char *str, const char *delim, char **context);
#define SUPERFRAME_SLOTS (8)
#define SUPERFRAME_BUFFER_SIZE (SUPERFRAME_SLOTS * sizeof(uint32_t) + 2)
#define OPTION_BUFFER_SIZE 256
+#define COMPONENTS 4 // psnr & sse statistics maintained for total, y, u, v
static const char *DEFAULT_QUANTIZER_VALUES = "60,53,39,33,27";
static const char *DEFAULT_SCALE_FACTORS = "4/16,5/16,7/16,11/16,16/16";
@@ -47,16 +49,20 @@ static const char *DEFAULT_SCALE_FACTORS = "4/16,5/16,7/16,11/16,16/16";
typedef struct SvcInternal {
char options[OPTION_BUFFER_SIZE]; // set by vpx_svc_set_options
char quantizers[OPTION_BUFFER_SIZE]; // set by vpx_svc_set_quantizers
+ char quantizers_keyframe[OPTION_BUFFER_SIZE]; // set by
+ // vpx_svc_set_quantizers
char scale_factors[OPTION_BUFFER_SIZE]; // set by vpx_svc_set_scale_factors
// values extracted from option, quantizers
int scaling_factor_num[VPX_SS_MAX_LAYERS];
int scaling_factor_den[VPX_SS_MAX_LAYERS];
+ int quantizer_keyframe[VPX_SS_MAX_LAYERS];
int quantizer[VPX_SS_MAX_LAYERS];
// accumulated statistics
- double psnr_in_layer[VPX_SS_MAX_LAYERS];
- uint32_t bytes_in_layer[VPX_SS_MAX_LAYERS];
+ double psnr_sum[VPX_SS_MAX_LAYERS][COMPONENTS]; // total/Y/U/V
+ uint64_t sse_sum[VPX_SS_MAX_LAYERS][COMPONENTS];
+ uint32_t bytes_sum[VPX_SS_MAX_LAYERS];
// codec encoding values
int width; // width of highest layer
@@ -268,7 +274,8 @@ static vpx_codec_err_t set_option_encoding_mode(SvcContext *svc_ctx,
}
static vpx_codec_err_t parse_quantizer_values(SvcContext *svc_ctx,
- const char *quantizer_values) {
+ const char *quantizer_values,
+ const int is_keyframe) {
char *input_string;
char *token;
const char *delim = ",";
@@ -279,6 +286,11 @@ static vpx_codec_err_t parse_quantizer_values(SvcContext *svc_ctx,
SvcInternal *const si = get_svc_internal(svc_ctx);
if (quantizer_values == NULL || strlen(quantizer_values) == 0) {
+ if (is_keyframe) {
+ // If there non settings for key frame, we will apply settings from
+ // non key frame. So just simply return here.
+ return VPX_CODEC_INVALID_PARAM;
+ }
input_string = strdup(DEFAULT_QUANTIZER_VALUES);
} else {
input_string = strdup(quantizer_values);
@@ -299,7 +311,12 @@ static vpx_codec_err_t parse_quantizer_values(SvcContext *svc_ctx,
} else {
q = 0;
}
- si->quantizer[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] = q;
+ if (is_keyframe) {
+ si->quantizer_keyframe[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers]
+ = q;
+ } else {
+ si->quantizer[i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers] = q;
+ }
}
if (res == VPX_CODEC_OK && found != svc_ctx->spatial_layers) {
svc_log(svc_ctx, SVC_LOG_ERROR,
@@ -384,6 +401,7 @@ static vpx_codec_err_t parse_options(SvcContext *svc_ctx, const char *options) {
char *option_name;
char *option_value;
char *input_ptr;
+ int is_keyframe_qaunt_set = 0;
vpx_codec_err_t res = VPX_CODEC_OK;
if (options == NULL) return VPX_CODEC_OK;
@@ -409,8 +427,17 @@ static vpx_codec_err_t parse_options(SvcContext *svc_ctx, const char *options) {
res = parse_scale_factors(svc_ctx, option_value);
if (res != VPX_CODEC_OK) break;
} else if (strcmp("quantizers", option_name) == 0) {
- res = parse_quantizer_values(svc_ctx, option_value);
+ res = parse_quantizer_values(svc_ctx, option_value, 0);
+ if (res != VPX_CODEC_OK) break;
+ if (!is_keyframe_qaunt_set) {
+ SvcInternal *const si = get_svc_internal(svc_ctx);
+ memcpy(get_svc_internal(svc_ctx)->quantizer_keyframe, si->quantizer,
+ sizeof(si->quantizer));
+ }
+ } else if (strcmp("quantizers-keyframe", option_name) == 0) {
+ res = parse_quantizer_values(svc_ctx, option_value, 1);
if (res != VPX_CODEC_OK) break;
+ is_keyframe_qaunt_set = 1;
} else {
svc_log(svc_ctx, SVC_LOG_ERROR, "invalid option: %s\n", option_name);
res = VPX_CODEC_INVALID_PARAM;
@@ -433,13 +460,19 @@ vpx_codec_err_t vpx_svc_set_options(SvcContext *svc_ctx, const char *options) {
}
vpx_codec_err_t vpx_svc_set_quantizers(SvcContext *svc_ctx,
- const char *quantizers) {
+ const char *quantizers,
+ const int is_for_keyframe) {
SvcInternal *const si = get_svc_internal(svc_ctx);
if (svc_ctx == NULL || quantizers == NULL || si == NULL) {
return VPX_CODEC_INVALID_PARAM;
}
- strncpy(si->quantizers, quantizers, sizeof(si->quantizers));
- si->quantizers[sizeof(si->quantizers) - 1] = '\0';
+ if (is_for_keyframe) {
+ strncpy(si->quantizers_keyframe, quantizers, sizeof(si->quantizers));
+ si->quantizers_keyframe[sizeof(si->quantizers_keyframe) - 1] = '\0';
+ } else {
+ strncpy(si->quantizers, quantizers, sizeof(si->quantizers));
+ si->quantizers[sizeof(si->quantizers) - 1] = '\0';
+ }
return VPX_CODEC_OK;
}
@@ -490,9 +523,13 @@ vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
// for first frame
si->layers = svc_ctx->spatial_layers;
- res = parse_quantizer_values(svc_ctx, si->quantizers);
+ res = parse_quantizer_values(svc_ctx, si->quantizers, 0);
if (res != VPX_CODEC_OK) return res;
+ res = parse_quantizer_values(svc_ctx, si->quantizers_keyframe, 1);
+ if (res != VPX_CODEC_OK)
+ memcpy(si->quantizer_keyframe, si->quantizer, sizeof(si->quantizer));
+
res = parse_scale_factors(svc_ctx, si->scale_factors);
if (res != VPX_CODEC_OK) return res;
@@ -500,6 +537,34 @@ vpx_codec_err_t vpx_svc_init(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
res = parse_options(svc_ctx, si->options);
if (res != VPX_CODEC_OK) return res;
+ // Assign target bitrate for each layer. We calculate the ratio
+ // from the resolution for now.
+ // TODO(Minghai): Optimize the mechanism of allocating bits after
+ // implementing svc two pass rate control.
+ if (si->layers > 1) {
+ int i;
+ float total = 0;
+ float alloc_ratio[VPX_SS_MAX_LAYERS] = {0};
+
+ for (i = 0; i < si->layers; ++i) {
+ int pos = i + VPX_SS_MAX_LAYERS - svc_ctx->spatial_layers;
+ if (pos < VPX_SS_MAX_LAYERS && si->scaling_factor_den[pos] > 0) {
+ alloc_ratio[i] = (float)(si->scaling_factor_num[pos] * 1.0 /
+ si->scaling_factor_den[pos]);
+
+ alloc_ratio[i] *= alloc_ratio[i];
+ total += alloc_ratio[i];
+ }
+ }
+
+ for (i = 0; i < si->layers; ++i) {
+ if (total > 0) {
+ enc_cfg->ss_target_bitrate[i] = (unsigned int)
+ (enc_cfg->rc_target_bitrate * alloc_ratio[i] / total);
+ }
+ }
+ }
+
// modify encoder configuration
enc_cfg->ss_number_layers = si->layers;
enc_cfg->ts_number_layers = 1; // Temporal layers not used in this encoder.
@@ -713,8 +778,15 @@ static void set_svc_parameters(SvcContext *svc_ctx,
svc_log(svc_ctx, SVC_LOG_ERROR, "vpx_svc_get_layer_resolution failed\n");
}
layer_index = layer + VPX_SS_MAX_LAYERS - si->layers;
- svc_params.min_quantizer = si->quantizer[layer_index];
- svc_params.max_quantizer = si->quantizer[layer_index];
+
+ if (vpx_svc_is_keyframe(svc_ctx)) {
+ svc_params.min_quantizer = si->quantizer_keyframe[layer_index];
+ svc_params.max_quantizer = si->quantizer_keyframe[layer_index];
+ } else {
+ svc_params.min_quantizer = si->quantizer[layer_index];
+ svc_params.max_quantizer = si->quantizer[layer_index];
+ }
+
svc_params.distance_from_i_frame = si->frame_within_gop;
// Use buffer i for layer i LST
@@ -814,7 +886,7 @@ vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
switch (cx_pkt->kind) {
case VPX_CODEC_CX_FRAME_PKT: {
const uint32_t frame_pkt_size = (uint32_t)(cx_pkt->data.frame.sz);
- si->bytes_in_layer[si->layer] += frame_pkt_size;
+ si->bytes_sum[si->layer] += frame_pkt_size;
svc_log(svc_ctx, SVC_LOG_DEBUG,
"SVC frame: %d, layer: %d, size: %u\n",
si->encode_frame_count, si->layer, frame_pkt_size);
@@ -832,13 +904,23 @@ vpx_codec_err_t vpx_svc_encode(SvcContext *svc_ctx, vpx_codec_ctx_t *codec_ctx,
break;
}
case VPX_CODEC_PSNR_PKT: {
+ int i;
svc_log(svc_ctx, SVC_LOG_DEBUG,
"SVC frame: %d, layer: %d, PSNR(Total/Y/U/V): "
"%2.3f %2.3f %2.3f %2.3f \n",
si->encode_frame_count, si->layer,
cx_pkt->data.psnr.psnr[0], cx_pkt->data.psnr.psnr[1],
cx_pkt->data.psnr.psnr[2], cx_pkt->data.psnr.psnr[3]);
- si->psnr_in_layer[si->layer] += cx_pkt->data.psnr.psnr[0];
+ svc_log(svc_ctx, SVC_LOG_DEBUG,
+ "SVC frame: %d, layer: %d, SSE(Total/Y/U/V): "
+ "%2.3f %2.3f %2.3f %2.3f \n",
+ si->encode_frame_count, si->layer,
+ cx_pkt->data.psnr.sse[0], cx_pkt->data.psnr.sse[1],
+ cx_pkt->data.psnr.sse[2], cx_pkt->data.psnr.sse[3]);
+ for (i = 0; i < COMPONENTS; i++) {
+ si->psnr_sum[si->layer][i] += cx_pkt->data.psnr.psnr[i];
+ si->sse_sum[si->layer][i] += cx_pkt->data.psnr.sse[i];
+ }
break;
}
default: {
@@ -916,11 +998,21 @@ void vpx_svc_set_keyframe(SvcContext *svc_ctx) {
si->frame_within_gop = 0;
}
+static double calc_psnr(double d) {
+ if (d == 0) return 100;
+ return -10.0 * log(d) / log(10.0);
+}
+
// dump accumulated statistics and reset accumulated values
const char *vpx_svc_dump_statistics(SvcContext *svc_ctx) {
int number_of_frames, number_of_keyframes, encode_frame_count;
- int i;
+ int i, j;
uint32_t bytes_total = 0;
+ double scale[COMPONENTS];
+ double psnr[COMPONENTS];
+ double mse[COMPONENTS];
+ double y_scale;
+
SvcInternal *const si = get_svc_internal(svc_ctx);
if (svc_ctx == NULL || si == NULL) return NULL;
@@ -938,12 +1030,36 @@ const char *vpx_svc_dump_statistics(SvcContext *svc_ctx) {
(i == 1 || i == 3)) {
number_of_frames -= number_of_keyframes;
}
- svc_log(svc_ctx, SVC_LOG_INFO, "Layer %d PSNR=[%2.3f], Bytes=[%u]\n", i,
- (double)si->psnr_in_layer[i] / number_of_frames,
- si->bytes_in_layer[i]);
- bytes_total += si->bytes_in_layer[i];
- si->psnr_in_layer[i] = 0;
- si->bytes_in_layer[i] = 0;
+ svc_log(svc_ctx, SVC_LOG_INFO,
+ "Layer %d Average PSNR=[%2.3f, %2.3f, %2.3f, %2.3f], Bytes=[%u]\n",
+ i, (double)si->psnr_sum[i][0] / number_of_frames,
+ (double)si->psnr_sum[i][1] / number_of_frames,
+ (double)si->psnr_sum[i][2] / number_of_frames,
+ (double)si->psnr_sum[i][3] / number_of_frames, si->bytes_sum[i]);
+ // the following psnr calculation is deduced from ffmpeg.c#print_report
+ y_scale = si->width * si->height * 255.0 * 255.0 * number_of_frames;
+ scale[1] = y_scale;
+ scale[2] = scale[3] = y_scale / 4; // U or V
+ scale[0] = y_scale * 1.5; // total
+
+ for (j = 0; j < COMPONENTS; j++) {
+ psnr[j] = calc_psnr(si->sse_sum[i][j] / scale[j]);
+ mse[j] = si->sse_sum[i][j] * 255.0 * 255.0 / scale[j];
+ }
+ svc_log(svc_ctx, SVC_LOG_INFO,
+ "Layer %d Overall PSNR=[%2.3f, %2.3f, %2.3f, %2.3f]\n", i, psnr[0],
+ psnr[1], psnr[2], psnr[3]);
+ svc_log(svc_ctx, SVC_LOG_INFO,
+ "Layer %d Overall MSE=[%2.3f, %2.3f, %2.3f, %2.3f]\n", i, mse[0],
+ mse[1], mse[2], mse[3]);
+
+ bytes_total += si->bytes_sum[i];
+ // clear sums for next time
+ si->bytes_sum[i] = 0;
+ for (j = 0; j < COMPONENTS; ++j) {
+ si->psnr_sum[i][j] = 0;
+ si->sse_sum[i][j] = 0;
+ }
}
// only display statistics once
diff --git a/vpx/src/vpx_encoder.c b/vpx/src/vpx_encoder.c
index 23742c8e8..e69d96efb 100644
--- a/vpx/src/vpx_encoder.c
+++ b/vpx/src/vpx_encoder.c
@@ -255,8 +255,8 @@ vpx_codec_err_t vpx_codec_encode(vpx_codec_ctx_t *ctx,
}
-const vpx_codec_cx_pkt_t *vpx_codec_get_cx_data(vpx_codec_ctx_t *ctx,
- vpx_codec_iter_t *iter) {
+const vpx_codec_cx_pkt_t *vpx_codec_get_cx_data(vpx_codec_ctx_t *ctx,
+ vpx_codec_iter_t *iter) {
const vpx_codec_cx_pkt_t *pkt = NULL;
if (ctx) {
@@ -271,32 +271,30 @@ const vpx_codec_cx_pkt_t *vpx_codec_get_cx_data(vpx_codec_ctx_t *ctx,
}
if (pkt && pkt->kind == VPX_CODEC_CX_FRAME_PKT) {
- /* If the application has specified a destination area for the
- * compressed data, and the codec has not placed the data there,
- * and it fits, copy it.
- */
- char *dst_buf = ctx->priv->enc.cx_data_dst_buf.buf;
-
- if (dst_buf
- && pkt->data.raw.buf != dst_buf
- && pkt->data.raw.sz
- + ctx->priv->enc.cx_data_pad_before
- + ctx->priv->enc.cx_data_pad_after
- <= ctx->priv->enc.cx_data_dst_buf.sz) {
- vpx_codec_cx_pkt_t *modified_pkt = &ctx->priv->enc.cx_data_pkt;
-
- memcpy(dst_buf + ctx->priv->enc.cx_data_pad_before,
- pkt->data.raw.buf, pkt->data.raw.sz);
+ // If the application has specified a destination area for the
+ // compressed data, and the codec has not placed the data there,
+ // and it fits, copy it.
+ vpx_codec_priv_t *const priv = ctx->priv;
+ char *const dst_buf = (char *)priv->enc.cx_data_dst_buf.buf;
+
+ if (dst_buf &&
+ pkt->data.raw.buf != dst_buf &&
+ pkt->data.raw.sz + priv->enc.cx_data_pad_before +
+ priv->enc.cx_data_pad_after <= priv->enc.cx_data_dst_buf.sz) {
+ vpx_codec_cx_pkt_t *modified_pkt = &priv->enc.cx_data_pkt;
+
+ memcpy(dst_buf + priv->enc.cx_data_pad_before, pkt->data.raw.buf,
+ pkt->data.raw.sz);
*modified_pkt = *pkt;
modified_pkt->data.raw.buf = dst_buf;
- modified_pkt->data.raw.sz += ctx->priv->enc.cx_data_pad_before
- + ctx->priv->enc.cx_data_pad_after;
+ modified_pkt->data.raw.sz += priv->enc.cx_data_pad_before +
+ priv->enc.cx_data_pad_after;
pkt = modified_pkt;
}
if (dst_buf == pkt->data.raw.buf) {
- ctx->priv->enc.cx_data_dst_buf.buf = dst_buf + pkt->data.raw.sz;
- ctx->priv->enc.cx_data_dst_buf.sz -= pkt->data.raw.sz;
+ priv->enc.cx_data_dst_buf.buf = dst_buf + pkt->data.raw.sz;
+ priv->enc.cx_data_dst_buf.sz -= pkt->data.raw.sz;
}
}
diff --git a/vpx/src/vpx_psnr.c b/vpx/src/vpx_psnr.c
new file mode 100644
index 000000000..05843acb6
--- /dev/null
+++ b/vpx/src/vpx_psnr.c
@@ -0,0 +1,24 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <math.h>
+
+#include "vpx/internal/vpx_psnr.h"
+
+#define MAX_PSNR 100.0
+
+double vpx_sse_to_psnr(double samples, double peak, double sse) {
+ if (sse > 0.0) {
+ const double psnr = 10.0 * log10(samples * peak * peak / sse);
+ return psnr > MAX_PSNR ? MAX_PSNR : psnr;
+ } else {
+ return MAX_PSNR;
+ }
+}
diff --git a/vpx/svc_context.h b/vpx/svc_context.h
index f675fb684..98474ca91 100644
--- a/vpx/svc_context.h
+++ b/vpx/svc_context.h
@@ -64,7 +64,8 @@ vpx_codec_err_t vpx_svc_set_options(SvcContext *svc_ctx, const char *options);
* e.g., "60,53,39,33,27"
*/
vpx_codec_err_t vpx_svc_set_quantizers(SvcContext *svc_ctx,
- const char *quantizer_values);
+ const char *quantizer_values,
+ const int is_for_keyframe);
/**
* Set SVC scale factors
diff --git a/vpx/vpx_codec.mk b/vpx/vpx_codec.mk
index 111c87e53..98d1d567c 100644
--- a/vpx/vpx_codec.mk
+++ b/vpx/vpx_codec.mk
@@ -34,8 +34,10 @@ API_SRCS-yes += vpx_decoder.h
API_SRCS-yes += src/vpx_encoder.c
API_SRCS-yes += vpx_encoder.h
API_SRCS-yes += internal/vpx_codec_internal.h
+API_SRCS-yes += internal/vpx_psnr.h
API_SRCS-yes += src/vpx_codec.c
API_SRCS-yes += src/vpx_image.c
+API_SRCS-yes += src/vpx_psnr.c
API_SRCS-yes += vpx_codec.h
API_SRCS-yes += vpx_codec.mk
API_SRCS-yes += vpx_frame_buffer.h
diff --git a/vpx/vpx_encoder.h b/vpx/vpx_encoder.h
index 1d9f0c9b7..851ff1ae8 100644
--- a/vpx/vpx_encoder.h
+++ b/vpx/vpx_encoder.h
@@ -610,6 +610,13 @@ extern "C" {
*/
unsigned int ss_number_layers;
+ /*!\brief Target bitrate for each spatial layer.
+ *
+ * These values specify the target coding bitrate to be used for each
+ * spatial layer.
+ */
+ unsigned int ss_target_bitrate[VPX_SS_MAX_LAYERS];
+
/*!\brief Number of temporal coding layers.
*
* This value specifies the number of temporal layers to be used.
generated by cgit v1.2.3 (git 2.25.1) at 2024-10-12 02:37:10 +0000