Clang format of video_processing folder.

BUG=webrtc:5259

Review URL: https://codereview.webrtc.org/1508793002

Cr-Commit-Position: refs/heads/master@{#10925}
This commit is contained in:
mflodman
2015-12-07 22:54:50 -08:00
committed by Commit bot
parent a440c6fa80
commit 99ab9447d1
27 changed files with 501 additions and 511 deletions

View File

@ -66,8 +66,8 @@ int32_t VPMBrightnessDetection::ProcessFrame(
for (int h = 0; h < height; h += (1 << stats.sub_sampling_factor)) { for (int h = 0; h < height; h += (1 << stats.sub_sampling_factor)) {
int row = h * width; int row = h * width;
for (int w = 0; w < width; w += (1 << stats.sub_sampling_factor)) { for (int w = 0; w < width; w += (1 << stats.sub_sampling_factor)) {
std_y += (buffer[w + row] - stats.mean) * (buffer[w + row] - std_y +=
stats.mean); (buffer[w + row] - stats.mean) * (buffer[w + row] - stats.mean);
} }
} }
std_y = sqrt(std_y / stats.num_pixels); std_y = sqrt(std_y / stats.num_pixels);
@ -82,8 +82,10 @@ int32_t VPMBrightnessDetection::ProcessFrame(
float posPerc95 = stats.num_pixels * 0.95f; float posPerc95 = stats.num_pixels * 0.95f;
for (uint32_t i = 0; i < 256; i++) { for (uint32_t i = 0; i < 256; i++) {
sum += stats.hist[i]; sum += stats.hist[i];
if (sum < pos_perc05) perc05 = i; // 5th perc. if (sum < pos_perc05)
if (sum < pos_median) median_y = i; // 50th perc. perc05 = i; // 5th perc.
if (sum < pos_median)
median_y = i; // 50th perc.
if (sum < posPerc95) if (sum < posPerc95)
perc95 = i; // 95th perc. perc95 = i; // 95th perc.
else else

View File

@ -135,12 +135,12 @@ int32_t VPMContentAnalysis::Initialize(int width, int height) {
} }
prev_frame_ = new uint8_t[width_ * height_]; // Y only. prev_frame_ = new uint8_t[width_ * height_]; // Y only.
if (prev_frame_ == NULL) return VPM_MEMORY; if (prev_frame_ == NULL)
return VPM_MEMORY;
return VPM_OK; return VPM_OK;
} }
// Compute motion metrics: magnitude over non-zero motion vectors, // Compute motion metrics: magnitude over non-zero motion vectors,
// and size of zero cluster // and size of zero cluster
int32_t VPMContentAnalysis::ComputeMotionMetrics() { int32_t VPMContentAnalysis::ComputeMotionMetrics() {
@ -173,21 +173,26 @@ int32_t VPMContentAnalysis::TemporalDiffMetric_C() {
uint8_t currPixel = orig_frame_[ssn]; uint8_t currPixel = orig_frame_[ssn];
uint8_t prevPixel = prev_frame_[ssn]; uint8_t prevPixel = prev_frame_[ssn];
tempDiffSum += (uint32_t)abs((int16_t)(currPixel - prevPixel)); tempDiffSum +=
pixelSum += (uint32_t) currPixel; static_cast<uint32_t>(abs((int16_t)(currPixel - prevPixel)));
pixelSqSum += (uint64_t) (currPixel * currPixel); pixelSum += static_cast<uint32_t>(currPixel);
pixelSqSum += static_cast<uint64_t>(currPixel * currPixel);
} }
} }
// Default. // Default.
motion_magnitude_ = 0.0f; motion_magnitude_ = 0.0f;
if (tempDiffSum == 0) return VPM_OK; if (tempDiffSum == 0)
return VPM_OK;
// Normalize over all pixels. // Normalize over all pixels.
float const tempDiffAvg = (float)tempDiffSum / (float)(num_pixels); float const tempDiffAvg =
float const pixelSumAvg = (float)pixelSum / (float)(num_pixels); static_cast<float>(tempDiffSum) / static_cast<float>(num_pixels);
float const pixelSqSumAvg = (float)pixelSqSum / (float)(num_pixels); float const pixelSumAvg =
static_cast<float>(pixelSum) / static_cast<float>(num_pixels);
float const pixelSqSumAvg =
static_cast<float>(pixelSqSum) / static_cast<float>(num_pixels);
float contrast = pixelSqSumAvg - (pixelSumAvg * pixelSumAvg); float contrast = pixelSqSumAvg - (pixelSumAvg * pixelSumAvg);
if (contrast > 0.0) { if (contrast > 0.0) {
@ -234,21 +239,22 @@ int32_t VPMContentAnalysis::ComputeSpatialMetrics_C() {
uint8_t rightPixel = orig_frame_[ssn4]; uint8_t rightPixel = orig_frame_[ssn4];
uint8_t leftPixel = orig_frame_[ssn5]; uint8_t leftPixel = orig_frame_[ssn5];
spatialErrSum += (uint32_t) abs((int16_t)(refPixel2 spatialErrSum += static_cast<uint32_t>(abs(static_cast<int16_t>(
- (uint16_t)(bottPixel + topPixel + leftPixel + rightPixel))); refPixel2 - static_cast<uint16_t>(bottPixel + topPixel + leftPixel +
spatialErrVSum += (uint32_t) abs((int16_t)(refPixel1 rightPixel))));
- (uint16_t)(bottPixel + topPixel))); spatialErrVSum += static_cast<uint32_t>(abs(static_cast<int16_t>(
spatialErrHSum += (uint32_t) abs((int16_t)(refPixel1 refPixel1 - static_cast<uint16_t>(bottPixel + topPixel))));
- (uint16_t)(leftPixel + rightPixel))); spatialErrHSum += static_cast<uint32_t>(abs(static_cast<int16_t>(
refPixel1 - static_cast<uint16_t>(leftPixel + rightPixel))));
pixelMSA += orig_frame_[ssn1]; pixelMSA += orig_frame_[ssn1];
} }
} }
// Normalize over all pixels. // Normalize over all pixels.
const float spatialErr = (float)(spatialErrSum >> 2); const float spatialErr = static_cast<float>(spatialErrSum >> 2);
const float spatialErrH = (float)(spatialErrHSum >> 1); const float spatialErrH = static_cast<float>(spatialErrHSum >> 1);
const float spatialErrV = (float)(spatialErrVSum >> 1); const float spatialErrV = static_cast<float>(spatialErrVSum >> 1);
const float norm = (float)pixelMSA; const float norm = static_cast<float>(pixelMSA);
// 2X2: // 2X2:
spatial_pred_err_ = spatialErr / norm; spatial_pred_err_ = spatialErr / norm;
@ -260,7 +266,8 @@ int32_t VPMContentAnalysis::ComputeSpatialMetrics_C() {
} }
VideoContentMetrics* VPMContentAnalysis::ContentMetrics() { VideoContentMetrics* VPMContentAnalysis::ContentMetrics() {
if (ca_Init_ == false) return NULL; if (ca_Init_ == false)
return NULL;
content_metrics_->spatial_pred_err = spatial_pred_err_; content_metrics_->spatial_pred_err = spatial_pred_err_;
content_metrics_->spatial_pred_err_h = spatial_pred_err_h_; content_metrics_->spatial_pred_err_h = spatial_pred_err_h_;

View File

@ -66,8 +66,8 @@ int32_t VPMContentAnalysis::TemporalDiffMetric_SSE2() {
} }
// Add to 64 bit running sum as to not roll over. // Add to 64 bit running sum as to not roll over.
sqsum_64 = _mm_add_epi64(sqsum_64, sqsum_64 =
_mm_add_epi64(_mm_unpackhi_epi32(sqsum_32,z), _mm_add_epi64(sqsum_64, _mm_add_epi64(_mm_unpackhi_epi32(sqsum_32, z),
_mm_unpacklo_epi32(sqsum_32, z))); _mm_unpacklo_epi32(sqsum_32, z)));
imgBufO += width_ * skip_num_; imgBufO += width_ * skip_num_;
@ -96,12 +96,16 @@ int32_t VPMContentAnalysis::TemporalDiffMetric_SSE2() {
// Default. // Default.
motion_magnitude_ = 0.0f; motion_magnitude_ = 0.0f;
if (tempDiffSum == 0) return VPM_OK; if (tempDiffSum == 0)
return VPM_OK;
// Normalize over all pixels. // Normalize over all pixels.
const float tempDiffAvg = (float)tempDiffSum / (float)(num_pixels); const float tempDiffAvg =
const float pixelSumAvg = (float)pixelSum / (float)(num_pixels); static_cast<float>(tempDiffSum) / static_cast<float>(num_pixels);
const float pixelSqSumAvg = (float)pixelSqSum / (float)(num_pixels); const float pixelSumAvg =
static_cast<float>(pixelSum) / static_cast<float>(num_pixels);
const float pixelSqSumAvg =
static_cast<float>(pixelSqSum) / static_cast<float>(num_pixels);
float contrast = pixelSqSumAvg - (pixelSumAvg * pixelSumAvg); float contrast = pixelSqSumAvg - (pixelSumAvg * pixelSumAvg);
if (contrast > 0.0) { if (contrast > 0.0) {
@ -163,16 +167,16 @@ int32_t VPMContentAnalysis::ComputeSpatialMetrics_SSE2() {
__m128i chi = _mm_unpackhi_epi8(c, z); __m128i chi = _mm_unpackhi_epi8(c, z);
// left right pixels unpacked and added together // left right pixels unpacked and added together
const __m128i lrlo = _mm_add_epi16(_mm_unpacklo_epi8(l,z), const __m128i lrlo =
_mm_unpacklo_epi8(r,z)); _mm_add_epi16(_mm_unpacklo_epi8(l, z), _mm_unpacklo_epi8(r, z));
const __m128i lrhi = _mm_add_epi16(_mm_unpackhi_epi8(l,z), const __m128i lrhi =
_mm_unpackhi_epi8(r,z)); _mm_add_epi16(_mm_unpackhi_epi8(l, z), _mm_unpackhi_epi8(r, z));
// top & bottom pixels unpacked and added together // top & bottom pixels unpacked and added together
const __m128i tblo = _mm_add_epi16(_mm_unpacklo_epi8(t,z), const __m128i tblo =
_mm_unpacklo_epi8(b,z)); _mm_add_epi16(_mm_unpacklo_epi8(t, z), _mm_unpacklo_epi8(b, z));
const __m128i tbhi = _mm_add_epi16(_mm_unpackhi_epi8(t,z), const __m128i tbhi =
_mm_unpackhi_epi8(b,z)); _mm_add_epi16(_mm_unpackhi_epi8(t, z), _mm_unpackhi_epi8(b, z));
// running sum of all pixels // running sum of all pixels
msa_16 = _mm_add_epi16(msa_16, _mm_add_epi16(chi, clo)); msa_16 = _mm_add_epi16(msa_16, _mm_add_epi16(chi, clo));
@ -190,28 +194,31 @@ int32_t VPMContentAnalysis::ComputeSpatialMetrics_SSE2() {
const __m128i sethi = _mm_subs_epi16(chi, _mm_add_epi16(lrhi, tbhi)); const __m128i sethi = _mm_subs_epi16(chi, _mm_add_epi16(lrhi, tbhi));
// Add to 16 bit running sum // Add to 16 bit running sum
se_16 = _mm_add_epi16(se_16, _mm_max_epi16(setlo, se_16 =
_mm_subs_epi16(z, setlo))); _mm_add_epi16(se_16, _mm_max_epi16(setlo, _mm_subs_epi16(z, setlo)));
se_16 = _mm_add_epi16(se_16, _mm_max_epi16(sethi, se_16 =
_mm_subs_epi16(z, sethi))); _mm_add_epi16(se_16, _mm_max_epi16(sethi, _mm_subs_epi16(z, sethi)));
sev_16 = _mm_add_epi16(sev_16, _mm_max_epi16(sevtlo, sev_16 = _mm_add_epi16(sev_16,
_mm_subs_epi16(z, sevtlo))); _mm_max_epi16(sevtlo, _mm_subs_epi16(z, sevtlo)));
sev_16 = _mm_add_epi16(sev_16, _mm_max_epi16(sevthi, sev_16 = _mm_add_epi16(sev_16,
_mm_subs_epi16(z, sevthi))); _mm_max_epi16(sevthi, _mm_subs_epi16(z, sevthi)));
seh_16 = _mm_add_epi16(seh_16, _mm_max_epi16(sehtlo, seh_16 = _mm_add_epi16(seh_16,
_mm_subs_epi16(z, sehtlo))); _mm_max_epi16(sehtlo, _mm_subs_epi16(z, sehtlo)));
seh_16 = _mm_add_epi16(seh_16, _mm_max_epi16(sehthi, seh_16 = _mm_add_epi16(seh_16,
_mm_subs_epi16(z, sehthi))); _mm_max_epi16(sehthi, _mm_subs_epi16(z, sehthi)));
} }
// Add to 32 bit running sum as to not roll over. // Add to 32 bit running sum as to not roll over.
se_32 = _mm_add_epi32(se_32, _mm_add_epi32(_mm_unpackhi_epi16(se_16, z), se_32 = _mm_add_epi32(se_32, _mm_add_epi32(_mm_unpackhi_epi16(se_16, z),
_mm_unpacklo_epi16(se_16, z))); _mm_unpacklo_epi16(se_16, z)));
sev_32 = _mm_add_epi32(sev_32, _mm_add_epi32(_mm_unpackhi_epi16(sev_16,z), sev_32 =
_mm_add_epi32(sev_32, _mm_add_epi32(_mm_unpackhi_epi16(sev_16, z),
_mm_unpacklo_epi16(sev_16, z))); _mm_unpacklo_epi16(sev_16, z)));
seh_32 = _mm_add_epi32(seh_32, _mm_add_epi32(_mm_unpackhi_epi16(seh_16,z), seh_32 =
_mm_add_epi32(seh_32, _mm_add_epi32(_mm_unpackhi_epi16(seh_16, z),
_mm_unpacklo_epi16(seh_16, z))); _mm_unpacklo_epi16(seh_16, z)));
msa_32 = _mm_add_epi32(msa_32, _mm_add_epi32(_mm_unpackhi_epi16(msa_16,z), msa_32 =
_mm_add_epi32(msa_32, _mm_add_epi32(_mm_unpackhi_epi16(msa_16, z),
_mm_unpacklo_epi16(msa_16, z))); _mm_unpacklo_epi16(msa_16, z)));
imgBuf += width_ * skip_num_; imgBuf += width_ * skip_num_;
@ -244,10 +251,10 @@ int32_t VPMContentAnalysis::ComputeSpatialMetrics_SSE2() {
const uint32_t pixelMSA = msa_64[0] + msa_64[1]; const uint32_t pixelMSA = msa_64[0] + msa_64[1];
// Normalize over all pixels. // Normalize over all pixels.
const float spatialErr = (float)(spatialErrSum >> 2); const float spatialErr = static_cast<float>(spatialErrSum >> 2);
const float spatialErrH = (float)(spatialErrHSum >> 1); const float spatialErrH = static_cast<float>(spatialErrHSum >> 1);
const float spatialErrV = (float)(spatialErrVSum >> 1); const float spatialErrV = static_cast<float>(spatialErrVSum >> 1);
const float norm = (float)pixelMSA; const float norm = static_cast<float>(pixelMSA);
// 2X2: // 2X2:
spatial_pred_err_ = spatialErr / norm; spatial_pred_err_ = spatialErr / norm;

View File

@ -40,16 +40,17 @@ enum { kLog2OfDownsamplingFactor = 3 };
// >> fprintf('%d, ', probUW16) // >> fprintf('%d, ', probUW16)
// Resolution reduced to avoid overflow when multiplying with the // Resolution reduced to avoid overflow when multiplying with the
// (potentially) large number of pixels. // (potentially) large number of pixels.
const uint16_t VPMDeflickering::prob_uw16_[kNumProbs] = {102, 205, 410, 614, const uint16_t VPMDeflickering::prob_uw16_[kNumProbs] = {
819, 1024, 1229, 1434, 1638, 1843, 1946, 1987}; // <Q11> 102, 205, 410, 614, 819, 1024,
1229, 1434, 1638, 1843, 1946, 1987}; // <Q11>
// To generate in Matlab: // To generate in Matlab:
// >> numQuants = 14; maxOnlyLength = 5; // >> numQuants = 14; maxOnlyLength = 5;
// >> weightUW16 = round(2^15 * // >> weightUW16 = round(2^15 *
// [linspace(0.5, 1.0, numQuants - maxOnlyLength)]); // [linspace(0.5, 1.0, numQuants - maxOnlyLength)]);
// >> fprintf('%d, %d,\n ', weightUW16); // >> fprintf('%d, %d,\n ', weightUW16);
const uint16_t VPMDeflickering::weight_uw16_[kNumQuants - kMaxOnlyLength] = const uint16_t VPMDeflickering::weight_uw16_[kNumQuants - kMaxOnlyLength] = {
{16384, 18432, 20480, 22528, 24576, 26624, 28672, 30720, 32768}; // <Q15> 16384, 18432, 20480, 22528, 24576, 26624, 28672, 30720, 32768}; // <Q15>
VPMDeflickering::VPMDeflickering() { VPMDeflickering::VPMDeflickering() {
Reset(); Reset();
@ -70,8 +71,8 @@ void VPMDeflickering::Reset() {
quant_hist_uw8_[0][kNumQuants - 1] = 255; quant_hist_uw8_[0][kNumQuants - 1] = 255;
for (int32_t i = 0; i < kNumProbs; i++) { for (int32_t i = 0; i < kNumProbs; i++) {
// Unsigned round. <Q0> // Unsigned round. <Q0>
quant_hist_uw8_[0][i + 1] = static_cast<uint8_t>( quant_hist_uw8_[0][i + 1] =
(prob_uw16_[i] * 255 + (1 << 10)) >> 11); static_cast<uint8_t>((prob_uw16_[i] * 255 + (1 << 10)) >> 11);
} }
for (int32_t i = 1; i < kFrameHistory_size; i++) { for (int32_t i = 1; i < kFrameHistory_size; i++) {
@ -80,8 +81,7 @@ void VPMDeflickering::Reset() {
} }
} }
int32_t VPMDeflickering::ProcessFrame( int32_t VPMDeflickering::ProcessFrame(VideoFrame* frame,
VideoFrame* frame,
VideoProcessing::FrameStats* stats) { VideoProcessing::FrameStats* stats) {
assert(frame); assert(frame);
uint32_t frame_memory; uint32_t frame_memory;
@ -111,7 +111,8 @@ int32_t VPMDeflickering::ProcessFrame(
return VPM_GENERAL_ERROR; return VPM_GENERAL_ERROR;
} }
if (PreDetection(frame->timestamp(), *stats) == -1) return VPM_GENERAL_ERROR; if (PreDetection(frame->timestamp(), *stats) == -1)
return VPM_GENERAL_ERROR;
// Flicker detection // Flicker detection
int32_t det_flicker = DetectFlicker(); int32_t det_flicker = DetectFlicker();
@ -124,13 +125,13 @@ int32_t VPMDeflickering::ProcessFrame(
// Size of luminance component. // Size of luminance component.
const uint32_t y_size = height * width; const uint32_t y_size = height * width;
const uint32_t y_sub_size = width * (((height - 1) >> const uint32_t y_sub_size =
kLog2OfDownsamplingFactor) + 1); width * (((height - 1) >> kLog2OfDownsamplingFactor) + 1);
uint8_t* y_sorted = new uint8_t[y_sub_size]; uint8_t* y_sorted = new uint8_t[y_sub_size];
uint32_t sort_row_idx = 0; uint32_t sort_row_idx = 0;
for (int i = 0; i < height; i += kDownsamplingFactor) { for (int i = 0; i < height; i += kDownsamplingFactor) {
memcpy(y_sorted + sort_row_idx * width, memcpy(y_sorted + sort_row_idx * width, frame->buffer(kYPlane) + i * width,
frame->buffer(kYPlane) + i * width, width); width);
sort_row_idx++; sort_row_idx++;
} }
@ -190,9 +191,10 @@ int32_t VPMDeflickering::ProcessFrame(
// target = w * maxquant_uw8 + (1 - w) * minquant_uw8 // target = w * maxquant_uw8 + (1 - w) * minquant_uw8
// Weights w = |weight_uw16_| are in Q15, hence the final output has to be // Weights w = |weight_uw16_| are in Q15, hence the final output has to be
// right shifted by 8 to end up in Q7. // right shifted by 8 to end up in Q7.
target_quant_uw16[i] = static_cast<uint16_t>(( target_quant_uw16[i] = static_cast<uint16_t>(
weight_uw16_[i] * maxquant_uw8[i] + (weight_uw16_[i] * maxquant_uw8[i] +
((1 << 15) - weight_uw16_[i]) * minquant_uw8[i]) >> 8); // <Q7> ((1 << 15) - weight_uw16_[i]) * minquant_uw8[i]) >>
8); // <Q7>
} }
for (int32_t i = kNumQuants - kMaxOnlyLength; i < kNumQuants; i++) { for (int32_t i = kNumQuants - kMaxOnlyLength; i < kNumQuants; i++) {
@ -203,12 +205,13 @@ int32_t VPMDeflickering::ProcessFrame(
uint16_t mapUW16; // <Q7> uint16_t mapUW16; // <Q7>
for (int32_t i = 1; i < kNumQuants; i++) { for (int32_t i = 1; i < kNumQuants; i++) {
// As quant and targetQuant are limited to UWord8, it's safe to use Q7 here. // As quant and targetQuant are limited to UWord8, it's safe to use Q7 here.
tmp_uw32 = static_cast<uint32_t>(target_quant_uw16[i] - tmp_uw32 =
target_quant_uw16[i - 1]); static_cast<uint32_t>(target_quant_uw16[i] - target_quant_uw16[i - 1]);
tmp_uw16 = static_cast<uint16_t>(quant_uw8[i] - quant_uw8[i - 1]); // <Q0> tmp_uw16 = static_cast<uint16_t>(quant_uw8[i] - quant_uw8[i - 1]); // <Q0>
if (tmp_uw16 > 0) { if (tmp_uw16 > 0) {
increment_uw16 = static_cast<uint16_t>(WebRtcSpl_DivU32U16(tmp_uw32, increment_uw16 =
static_cast<uint16_t>(WebRtcSpl_DivU32U16(tmp_uw32,
tmp_uw16)); // <Q7> tmp_uw16)); // <Q7>
} else { } else {
// The value is irrelevant; the loop below will only iterate once. // The value is irrelevant; the loop below will only iterate once.
@ -247,7 +250,8 @@ int32_t VPMDeflickering::ProcessFrame(
zero.\n zero.\n
-1: Error -1: Error
*/ */
int32_t VPMDeflickering::PreDetection(const uint32_t timestamp, int32_t VPMDeflickering::PreDetection(
const uint32_t timestamp,
const VideoProcessing::FrameStats& stats) { const VideoProcessing::FrameStats& stats) {
int32_t mean_val; // Mean value of frame (Q4) int32_t mean_val; // Mean value of frame (Q4)
uint32_t frame_rate = 0; uint32_t frame_rate = 0;
@ -262,8 +266,8 @@ int32_t VPMDeflickering::PreDetection(const uint32_t timestamp,
// Update timestamp buffer. // Update timestamp buffer.
// This should be done even though we might end up in an unreliable detection. // This should be done even though we might end up in an unreliable detection.
memmove(timestamp_buffer_ + 1, timestamp_buffer_, (kMeanBufferLength - 1) * memmove(timestamp_buffer_ + 1, timestamp_buffer_,
sizeof(uint32_t)); (kMeanBufferLength - 1) * sizeof(uint32_t));
timestamp_buffer_[0] = timestamp; timestamp_buffer_[0] = timestamp;
/* Compute current frame rate (Q4) */ /* Compute current frame rate (Q4) */
@ -322,7 +326,7 @@ int32_t VPMDeflickering::DetectFlicker() {
/* Sanity check for mean_buffer_length_ */ /* Sanity check for mean_buffer_length_ */
if (mean_buffer_length_ < 2) { if (mean_buffer_length_ < 2) {
/* Not possible to estimate frequency */ /* Not possible to estimate frequency */
return(2); return 2;
} }
// Count zero crossings with a dead zone to be robust against noise. If the // Count zero crossings with a dead zone to be robust against noise. If the
// noise std is 2 pixel this corresponds to about 95% confidence interval. // noise std is 2 pixel this corresponds to about 95% confidence interval.

View File

@ -53,15 +53,17 @@ void VPMFramePreprocessor::SetInputFrameResampleMode(
spatial_resampler_->SetInputFrameResampleMode(resampling_mode); spatial_resampler_->SetInputFrameResampleMode(resampling_mode);
} }
int32_t VPMFramePreprocessor::SetTargetResolution( int32_t VPMFramePreprocessor::SetTargetResolution(uint32_t width,
uint32_t width, uint32_t height, uint32_t frame_rate) { uint32_t height,
uint32_t frame_rate) {
if ((width == 0) || (height == 0) || (frame_rate == 0)) { if ((width == 0) || (height == 0) || (frame_rate == 0)) {
return VPM_PARAMETER_ERROR; return VPM_PARAMETER_ERROR;
} }
int32_t ret_val = 0; int32_t ret_val = 0;
ret_val = spatial_resampler_->SetTargetFrameSize(width, height); ret_val = spatial_resampler_->SetTargetFrameSize(width, height);
if (ret_val < 0) return ret_val; if (ret_val < 0)
return ret_val;
vd_->SetTargetFramerate(frame_rate); vd_->SetTargetFramerate(frame_rate);
return VPM_OK; return VPM_OK;
@ -84,12 +86,10 @@ uint32_t VPMFramePreprocessor::GetDecimatedFrameRate() {
return vd_->GetDecimatedFrameRate(); return vd_->GetDecimatedFrameRate();
} }
uint32_t VPMFramePreprocessor::GetDecimatedWidth() const { uint32_t VPMFramePreprocessor::GetDecimatedWidth() const {
return spatial_resampler_->TargetWidth(); return spatial_resampler_->TargetWidth();
} }
uint32_t VPMFramePreprocessor::GetDecimatedHeight() const { uint32_t VPMFramePreprocessor::GetDecimatedHeight() const {
return spatial_resampler_->TargetHeight(); return spatial_resampler_->TargetHeight();
} }

View File

@ -41,7 +41,8 @@ class VPMFramePreprocessor {
void EnableContentAnalysis(bool enable); void EnableContentAnalysis(bool enable);
// Set target resolution: frame rate and dimension. // Set target resolution: frame rate and dimension.
int32_t SetTargetResolution(uint32_t width, uint32_t height, int32_t SetTargetResolution(uint32_t width,
uint32_t height,
uint32_t frame_rate); uint32_t frame_rate);
// Set target frame rate. // Set target frame rate.

View File

@ -8,13 +8,6 @@
* be found in the AUTHORS file in the root of the source tree. * be found in the AUTHORS file in the root of the source tree.
*/ */
/*
* video_processing.h
* This header file contains the API required for the video
* processing module class.
*/
#ifndef WEBRTC_MODULES_VIDEO_PROCESSING_INCLUDE_VIDEO_PROCESSING_H_ #ifndef WEBRTC_MODULES_VIDEO_PROCESSING_INCLUDE_VIDEO_PROCESSING_H_
#define WEBRTC_MODULES_VIDEO_PROCESSING_INCLUDE_VIDEO_PROCESSING_H_ #define WEBRTC_MODULES_VIDEO_PROCESSING_INCLUDE_VIDEO_PROCESSING_H_
@ -43,11 +36,7 @@ class VideoProcessing {
uint32_t sub_sampling_factor; // Sub-sampling factor, in powers of 2. uint32_t sub_sampling_factor; // Sub-sampling factor, in powers of 2.
}; };
enum BrightnessWarning { enum BrightnessWarning { kNoWarning, kDarkWarning, kBrightWarning };
kNoWarning,
kDarkWarning,
kBrightWarning
};
static VideoProcessing* Create(); static VideoProcessing* Create();
virtual ~VideoProcessing() {} virtual ~VideoProcessing() {}

View File

@ -10,7 +10,6 @@
#include "webrtc/modules/video_processing/spatial_resampler.h" #include "webrtc/modules/video_processing/spatial_resampler.h"
namespace webrtc { namespace webrtc {
VPMSimpleSpatialResampler::VPMSimpleSpatialResampler() VPMSimpleSpatialResampler::VPMSimpleSpatialResampler()
@ -21,12 +20,13 @@ VPMSimpleSpatialResampler::VPMSimpleSpatialResampler()
VPMSimpleSpatialResampler::~VPMSimpleSpatialResampler() {} VPMSimpleSpatialResampler::~VPMSimpleSpatialResampler() {}
int32_t VPMSimpleSpatialResampler::SetTargetFrameSize(int32_t width, int32_t VPMSimpleSpatialResampler::SetTargetFrameSize(int32_t width,
int32_t height) { int32_t height) {
if (resampling_mode_ == kNoRescaling) return VPM_OK; if (resampling_mode_ == kNoRescaling)
return VPM_OK;
if (width < 1 || height < 1) return VPM_PARAMETER_ERROR; if (width < 1 || height < 1)
return VPM_PARAMETER_ERROR;
target_width_ = width; target_width_ = width;
target_height_ = height; target_height_ = height;
@ -48,10 +48,10 @@ void VPMSimpleSpatialResampler::Reset() {
int32_t VPMSimpleSpatialResampler::ResampleFrame(const VideoFrame& inFrame, int32_t VPMSimpleSpatialResampler::ResampleFrame(const VideoFrame& inFrame,
VideoFrame* outFrame) { VideoFrame* outFrame) {
// Don't copy if frame remains as is. // Don't copy if frame remains as is.
if (resampling_mode_ == kNoRescaling) if (resampling_mode_ == kNoRescaling) {
return VPM_OK; return VPM_OK;
// Check if re-sampling is needed // Check if re-sampling is needed
else if ((inFrame.width() == target_width_) && } else if ((inFrame.width() == target_width_) &&
(inFrame.height() == target_height_)) { (inFrame.height() == target_height_)) {
return VPM_OK; return VPM_OK;
} }
@ -60,8 +60,8 @@ int32_t VPMSimpleSpatialResampler::ResampleFrame(const VideoFrame& inFrame,
// TODO(mikhal/marpan): Should we allow for setting the filter mode in // TODO(mikhal/marpan): Should we allow for setting the filter mode in
// _scale.Set() with |resampling_mode_|? // _scale.Set() with |resampling_mode_|?
int ret_val = 0; int ret_val = 0;
ret_val = scaler_.Set(inFrame.width(), inFrame.height(), ret_val = scaler_.Set(inFrame.width(), inFrame.height(), target_width_,
target_width_, target_height_, kI420, kI420, kScaleBox); target_height_, kI420, kI420, kScaleBox);
if (ret_val < 0) if (ret_val < 0)
return ret_val; return ret_val;
@ -86,8 +86,7 @@ int32_t VPMSimpleSpatialResampler::TargetWidth() {
return target_width_; return target_width_;
} }
bool VPMSimpleSpatialResampler::ApplyResample(int32_t width, bool VPMSimpleSpatialResampler::ApplyResample(int32_t width, int32_t height) {
int32_t height) {
if ((width == target_width_ && height == target_height_) || if ((width == target_width_ && height == target_height_) ||
resampling_mode_ == kNoRescaling) resampling_mode_ == kNoRescaling)
return false; return false;

View File

@ -23,10 +23,10 @@ namespace webrtc {
class VPMSpatialResampler { class VPMSpatialResampler {
public: public:
virtual ~VPMSpatialResampler() {}; virtual ~VPMSpatialResampler() {}
virtual int32_t SetTargetFrameSize(int32_t width, int32_t height) = 0; virtual int32_t SetTargetFrameSize(int32_t width, int32_t height) = 0;
virtual void SetInputFrameResampleMode(VideoFrameResampling virtual void SetInputFrameResampleMode(
resampling_mode) = 0; VideoFrameResampling resampling_mode) = 0;
virtual void Reset() = 0; virtual void Reset() = 0;
virtual int32_t ResampleFrame(const VideoFrame& inFrame, virtual int32_t ResampleFrame(const VideoFrame& inFrame,
VideoFrame* outFrame) = 0; VideoFrame* outFrame) = 0;
@ -49,7 +49,6 @@ class VPMSimpleSpatialResampler : public VPMSpatialResampler {
virtual bool ApplyResample(int32_t width, int32_t height); virtual bool ApplyResample(int32_t width, int32_t height);
private: private:
VideoFrameResampling resampling_mode_; VideoFrameResampling resampling_mode_;
int32_t target_width_; int32_t target_width_;
int32_t target_height_; int32_t target_height_;

View File

@ -22,14 +22,14 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(BrightnessDetection)) {
rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]); rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]);
while (fread(video_buffer.get(), 1, frame_length_, source_file_) == while (fread(video_buffer.get(), 1, frame_length_, source_file_) ==
frame_length_) { frame_length_) {
EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_,
height_, 0, kVideoRotation_0, &video_frame_)); 0, kVideoRotation_0, &video_frame_));
frameNum++; frameNum++;
VideoProcessing::FrameStats stats; VideoProcessing::FrameStats stats;
vp_->GetFrameStats(video_frame_, &stats); vp_->GetFrameStats(video_frame_, &stats);
EXPECT_GT(stats.num_pixels, 0u); EXPECT_GT(stats.num_pixels, 0u);
ASSERT_GE(brightnessWarning = vp_->BrightnessDetection(video_frame_, ASSERT_GE(brightnessWarning = vp_->BrightnessDetection(video_frame_, stats),
stats), 0); 0);
if (brightnessWarning != VideoProcessing::kNoWarning) { if (brightnessWarning != VideoProcessing::kNoWarning) {
warningCount++; warningCount++;
} }
@ -48,8 +48,8 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(BrightnessDetection)) {
while (fread(video_buffer.get(), 1, frame_length_, source_file_) == while (fread(video_buffer.get(), 1, frame_length_, source_file_) ==
frame_length_ && frame_length_ &&
frameNum < 300) { frameNum < 300) {
EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_,
height_, 0, kVideoRotation_0, &video_frame_)); 0, kVideoRotation_0, &video_frame_));
frameNum++; frameNum++;
uint8_t* frame = video_frame_.buffer(kYPlane); uint8_t* frame = video_frame_.buffer(kYPlane);
@ -65,8 +65,8 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(BrightnessDetection)) {
VideoProcessing::FrameStats stats; VideoProcessing::FrameStats stats;
vp_->GetFrameStats(video_frame_, &stats); vp_->GetFrameStats(video_frame_, &stats);
EXPECT_GT(stats.num_pixels, 0u); EXPECT_GT(stats.num_pixels, 0u);
ASSERT_GE(brightnessWarning = vp_->BrightnessDetection(video_frame_, ASSERT_GE(brightnessWarning = vp_->BrightnessDetection(video_frame_, stats),
stats), 0); 0);
EXPECT_NE(VideoProcessing::kDarkWarning, brightnessWarning); EXPECT_NE(VideoProcessing::kDarkWarning, brightnessWarning);
if (brightnessWarning == VideoProcessing::kBrightWarning) { if (brightnessWarning == VideoProcessing::kBrightWarning) {
warningCount++; warningCount++;
@ -83,9 +83,10 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(BrightnessDetection)) {
frameNum = 0; frameNum = 0;
warningCount = 0; warningCount = 0;
while (fread(video_buffer.get(), 1, frame_length_, source_file_) == while (fread(video_buffer.get(), 1, frame_length_, source_file_) ==
frame_length_ && frameNum < 300) { frame_length_ &&
EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, frameNum < 300) {
height_, 0, kVideoRotation_0, &video_frame_)); EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_,
0, kVideoRotation_0, &video_frame_));
frameNum++; frameNum++;
uint8_t* y_plane = video_frame_.buffer(kYPlane); uint8_t* y_plane = video_frame_.buffer(kYPlane);
@ -98,8 +99,8 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(BrightnessDetection)) {
VideoProcessing::FrameStats stats; VideoProcessing::FrameStats stats;
vp_->GetFrameStats(video_frame_, &stats); vp_->GetFrameStats(video_frame_, &stats);
EXPECT_GT(stats.num_pixels, 0u); EXPECT_GT(stats.num_pixels, 0u);
ASSERT_GE(brightnessWarning = vp_->BrightnessDetection(video_frame_, ASSERT_GE(brightnessWarning = vp_->BrightnessDetection(video_frame_, stats),
stats), 0); 0);
EXPECT_NE(VideoProcessing::kBrightWarning, brightnessWarning); EXPECT_NE(VideoProcessing::kBrightWarning, brightnessWarning);
if (brightnessWarning == VideoProcessing::kDarkWarning) { if (brightnessWarning == VideoProcessing::kDarkWarning) {
warningCount++; warningCount++;

View File

@ -26,8 +26,8 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(ContentAnalysis)) {
ca__sse.Initialize(width_, height_); ca__sse.Initialize(width_, height_);
rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]); rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]);
while (fread(video_buffer.get(), 1, frame_length_, source_file_) while (fread(video_buffer.get(), 1, frame_length_, source_file_) ==
== frame_length_) { frame_length_) {
// Using ConvertToI420 to add stride to the image. // Using ConvertToI420 to add stride to the image.
EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_, EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_,
0, kVideoRotation_0, &video_frame_)); 0, kVideoRotation_0, &video_frame_));

View File

@ -33,14 +33,14 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(Deflickering)) {
const std::string input_file = const std::string input_file =
webrtc::test::ResourcePath("deflicker_before_cif_short", "yuv"); webrtc::test::ResourcePath("deflicker_before_cif_short", "yuv");
source_file_ = fopen(input_file.c_str(), "rb"); source_file_ = fopen(input_file.c_str(), "rb");
ASSERT_TRUE(source_file_ != NULL) << ASSERT_TRUE(source_file_ != NULL) << "Cannot read input file: " << input_file
"Cannot read input file: " << input_file << "\n"; << "\n";
const std::string output_file = const std::string output_file =
webrtc::test::OutputPath() + "deflicker_output_cif_short.yuv"; webrtc::test::OutputPath() + "deflicker_output_cif_short.yuv";
FILE* deflickerFile = fopen(output_file.c_str(), "wb"); FILE* deflickerFile = fopen(output_file.c_str(), "wb");
ASSERT_TRUE(deflickerFile != NULL) << ASSERT_TRUE(deflickerFile != NULL)
"Could not open output file: " << output_file << "\n"; << "Could not open output file: " << output_file << "\n";
printf("\nRun time [us / frame]:\n"); printf("\nRun time [us / frame]:\n");
rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]); rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]);
@ -54,8 +54,7 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(Deflickering)) {
while (fread(video_buffer.get(), 1, frame_length_, source_file_) == while (fread(video_buffer.get(), 1, frame_length_, source_file_) ==
frame_length_) { frame_length_) {
frameNum++; frameNum++;
EXPECT_EQ( EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_,
0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_,
height_, 0, kVideoRotation_0, &video_frame_)); height_, 0, kVideoRotation_0, &video_frame_));
video_frame_.set_timestamp(timeStamp); video_frame_.set_timestamp(timeStamp);

View File

@ -79,8 +79,8 @@ void VideoProcessingTest::SetUp() {
const std::string video_file = const std::string video_file =
webrtc::test::ResourcePath("foreman_cif", "yuv"); webrtc::test::ResourcePath("foreman_cif", "yuv");
source_file_ = fopen(video_file.c_str(), "rb"); source_file_ = fopen(video_file.c_str(), "rb");
ASSERT_TRUE(source_file_ != NULL) << ASSERT_TRUE(source_file_ != NULL)
"Cannot read source file: " + video_file + "\n"; << "Cannot read source file: " + video_file + "\n";
} }
void VideoProcessingTest::TearDown() { void VideoProcessingTest::TearDown() {
@ -110,8 +110,8 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(HandleBadStats)) {
VideoProcessing::FrameStats stats; VideoProcessing::FrameStats stats;
vp_->ClearFrameStats(&stats); vp_->ClearFrameStats(&stats);
rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]); rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]);
ASSERT_EQ(frame_length_, fread(video_buffer.get(), 1, frame_length_, ASSERT_EQ(frame_length_,
source_file_)); fread(video_buffer.get(), 1, frame_length_, source_file_));
EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_, EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_,
0, kVideoRotation_0, &video_frame_)); 0, kVideoRotation_0, &video_frame_));
@ -125,8 +125,8 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(IdenticalResultsAfterReset)) {
VideoProcessing::FrameStats stats; VideoProcessing::FrameStats stats;
// Only testing non-static functions here. // Only testing non-static functions here.
rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]); rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]);
ASSERT_EQ(frame_length_, fread(video_buffer.get(), 1, frame_length_, ASSERT_EQ(frame_length_,
source_file_)); fread(video_buffer.get(), 1, frame_length_, source_file_));
EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_, EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_,
0, kVideoRotation_0, &video_frame_)); 0, kVideoRotation_0, &video_frame_));
vp_->GetFrameStats(video_frame_, &stats); vp_->GetFrameStats(video_frame_, &stats);
@ -140,8 +140,8 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(IdenticalResultsAfterReset)) {
ASSERT_EQ(0, vp_->Deflickering(&video_frame2, &stats)); ASSERT_EQ(0, vp_->Deflickering(&video_frame2, &stats));
EXPECT_TRUE(CompareFrames(video_frame_, video_frame2)); EXPECT_TRUE(CompareFrames(video_frame_, video_frame2));
ASSERT_EQ(frame_length_, fread(video_buffer.get(), 1, frame_length_, ASSERT_EQ(frame_length_,
source_file_)); fread(video_buffer.get(), 1, frame_length_, source_file_));
EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_, EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_,
0, kVideoRotation_0, &video_frame_)); 0, kVideoRotation_0, &video_frame_));
vp_->GetFrameStats(video_frame_, &stats); vp_->GetFrameStats(video_frame_, &stats);
@ -157,8 +157,8 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(FrameStats)) {
VideoProcessing::FrameStats stats; VideoProcessing::FrameStats stats;
vp_->ClearFrameStats(&stats); vp_->ClearFrameStats(&stats);
rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]); rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]);
ASSERT_EQ(frame_length_, fread(video_buffer.get(), 1, frame_length_, ASSERT_EQ(frame_length_,
source_file_)); fread(video_buffer.get(), 1, frame_length_, source_file_));
EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_, EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_,
0, kVideoRotation_0, &video_frame_)); 0, kVideoRotation_0, &video_frame_));
@ -204,8 +204,7 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(Resampler)) {
int64_t total_runtime = 0; int64_t total_runtime = 0;
rewind(source_file_); rewind(source_file_);
ASSERT_TRUE(source_file_ != NULL) << ASSERT_TRUE(source_file_ != NULL) << "Cannot read input file \n";
"Cannot read input file \n";
// CA not needed here // CA not needed here
vp_->EnableContentAnalysis(false); vp_->EnableContentAnalysis(false);
@ -214,8 +213,8 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(Resampler)) {
// Reading test frame // Reading test frame
rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]); rtc::scoped_ptr<uint8_t[]> video_buffer(new uint8_t[frame_length_]);
ASSERT_EQ(frame_length_, fread(video_buffer.get(), 1, frame_length_, ASSERT_EQ(frame_length_,
source_file_)); fread(video_buffer.get(), 1, frame_length_, source_file_));
// Using ConvertToI420 to add stride to the image. // Using ConvertToI420 to add stride to the image.
EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_, EXPECT_EQ(0, ConvertToI420(kI420, video_buffer.get(), 0, 0, width_, height_,
0, kVideoRotation_0, &video_frame_)); 0, kVideoRotation_0, &video_frame_));
@ -281,8 +280,7 @@ TEST_F(VideoProcessingTest, DISABLED_ON_IOS(Resampler)) {
printf("\nAverage run time = %d us / frame\n", printf("\nAverage run time = %d us / frame\n",
static_cast<int>(total_runtime)); static_cast<int>(total_runtime));
printf("Min run time = %d us / frame\n\n", printf("Min run time = %d us / frame\n\n", static_cast<int>(min_runtime));
static_cast<int>(min_runtime));
} }
void PreprocessFrameAndVerify(const VideoFrame& source, void PreprocessFrameAndVerify(const VideoFrame& source,
@ -349,10 +347,11 @@ void TestSize(const VideoFrame& source_frame,
// Compute PSNR against the cropped source frame and check expectation. // Compute PSNR against the cropped source frame and check expectation.
double psnr = I420PSNR(&cropped_source_frame, out_frame); double psnr = I420PSNR(&cropped_source_frame, out_frame);
EXPECT_GT(psnr, expected_psnr); EXPECT_GT(psnr, expected_psnr);
printf("PSNR: %f. PSNR is between source of size %d %d, and a modified " printf(
"PSNR: %f. PSNR is between source of size %d %d, and a modified "
"source which is scaled down/up to: %d %d, and back to source size \n", "source which is scaled down/up to: %d %d, and back to source size \n",
psnr, source_frame.width(), source_frame.height(), psnr, source_frame.width(), source_frame.height(), target_width,
target_width, target_height); target_height);
} }
bool CompareFrames(const webrtc::VideoFrame& frame1, bool CompareFrames(const webrtc::VideoFrame& frame1,

View File

@ -30,9 +30,7 @@ class VideoProcessingTest : public ::testing::Test {
std::string trace_file = webrtc::test::OutputPath() + "VPMTrace.txt"; std::string trace_file = webrtc::test::OutputPath() + "VPMTrace.txt";
ASSERT_EQ(0, Trace::SetTraceFile(trace_file.c_str())); ASSERT_EQ(0, Trace::SetTraceFile(trace_file.c_str()));
} }
static void TearDownTestCase() { static void TearDownTestCase() { Trace::ReturnTrace(); }
Trace::ReturnTrace();
}
VideoProcessing* vp_; VideoProcessing* vp_;
FILE* source_file_; FILE* source_file_;
VideoFrame video_frame_; VideoFrame video_frame_;

View File

@ -27,8 +27,7 @@ DenoiserFilter* DenoiserFilter::Create() {
#if defined(WEBRTC_ARCH_X86_FAMILY) #if defined(WEBRTC_ARCH_X86_FAMILY)
// x86 CPU detection required. // x86 CPU detection required.
if (WebRtc_GetCPUInfo(kSSE2)) { if (WebRtc_GetCPUInfo(kSSE2)) {
filter = filter = new DenoiserFilterSSE2();
new DenoiserFilterSSE2();
} else { } else {
filter = new DenoiserFilterC(); filter = new DenoiserFilterC();
} }

View File

@ -41,8 +41,6 @@ class DenoiserFilterC : public DenoiserFilter {
int increase_denoising) override; int increase_denoising) override;
}; };
} // namespace webrtc } // namespace webrtc
#endif // WEBRTC_MODULES_VIDEO_PROCESSING_UTIL_DENOISER_FILTER_C_H_ #endif // WEBRTC_MODULES_VIDEO_PROCESSING_UTIL_DENOISER_FILTER_C_H_

View File

@ -58,8 +58,8 @@ static void VarianceNeonW8(const uint8_t* a,
} }
*sum = HorizontalAddS16x8(v_sum); *sum = HorizontalAddS16x8(v_sum);
*sse = static_cast<uint32_t>( *sse =
HorizontalAddS32x4(vaddq_s32(v_sse_lo, v_sse_hi))); static_cast<uint32_t>(HorizontalAddS32x4(vaddq_s32(v_sse_lo, v_sse_hi)));
} }
void DenoiserFilterNEON::CopyMem16x16(const uint8_t* src, void DenoiserFilterNEON::CopyMem16x16(const uint8_t* src,
@ -111,8 +111,7 @@ DenoiserDecision DenoiserFilterNEON::MbDenoise(uint8_t* mc_running_avg_y,
// increasing the adjustment for each level, level1 adjustment is // increasing the adjustment for each level, level1 adjustment is
// increased, the deltas stay the same. // increased, the deltas stay the same.
int shift_inc = int shift_inc =
(increase_denoising && motion_magnitude <= kMotionMagnitudeThreshold) (increase_denoising && motion_magnitude <= kMotionMagnitudeThreshold) ? 1
? 1
: 0; : 0;
const uint8x16_t v_level1_adjustment = vmovq_n_u8( const uint8x16_t v_level1_adjustment = vmovq_n_u8(
(motion_magnitude <= kMotionMagnitudeThreshold) ? 4 + shift_inc : 3); (motion_magnitude <= kMotionMagnitudeThreshold) ? 4 + shift_inc : 3);

View File

@ -141,8 +141,7 @@ DenoiserDecision DenoiserFilterSSE2::MbDenoise(uint8_t* mc_running_avg_y,
uint8_t motion_magnitude, uint8_t motion_magnitude,
int increase_denoising) { int increase_denoising) {
int shift_inc = int shift_inc =
(increase_denoising && motion_magnitude <= kMotionMagnitudeThreshold) (increase_denoising && motion_magnitude <= kMotionMagnitudeThreshold) ? 1
? 1
: 0; : 0;
__m128i acc_diff = _mm_setzero_si128(); __m128i acc_diff = _mm_setzero_si128();
const __m128i k_0 = _mm_setzero_si128(); const __m128i k_0 = _mm_setzero_si128();
@ -239,15 +238,14 @@ DenoiserDecision DenoiserFilterSSE2::MbDenoise(uint8_t* mc_running_avg_y,
// Calculate differences. // Calculate differences.
const __m128i v_sig = const __m128i v_sig =
_mm_loadu_si128(reinterpret_cast<const __m128i*>(&sig[0])); _mm_loadu_si128(reinterpret_cast<const __m128i*>(&sig[0]));
const __m128i v_mc_running_avg_y = _mm_loadu_si128( const __m128i v_mc_running_avg_y =
reinterpret_cast<__m128i*>(&mc_running_avg_y[0])); _mm_loadu_si128(reinterpret_cast<__m128i*>(&mc_running_avg_y[0]));
const __m128i pdiff = _mm_subs_epu8(v_mc_running_avg_y, v_sig); const __m128i pdiff = _mm_subs_epu8(v_mc_running_avg_y, v_sig);
const __m128i ndiff = _mm_subs_epu8(v_sig, v_mc_running_avg_y); const __m128i ndiff = _mm_subs_epu8(v_sig, v_mc_running_avg_y);
// Obtain the sign. FF if diff is negative. // Obtain the sign. FF if diff is negative.
const __m128i diff_sign = _mm_cmpeq_epi8(pdiff, k_0); const __m128i diff_sign = _mm_cmpeq_epi8(pdiff, k_0);
// Clamp absolute difference to delta to get the adjustment. // Clamp absolute difference to delta to get the adjustment.
const __m128i adj = _mm_min_epu8( const __m128i adj = _mm_min_epu8(_mm_or_si128(pdiff, ndiff), k_delta);
_mm_or_si128(pdiff, ndiff), k_delta);
// Restore the sign and get positive and negative adjustments. // Restore the sign and get positive and negative adjustments.
__m128i padj, nadj; __m128i padj, nadj;
padj = _mm_andnot_si128(diff_sign, adj); padj = _mm_andnot_si128(diff_sign, adj);

18
webrtc/modules/video_processing/util/skin_detection.cc Executable file → Normal file
View File

@ -48,19 +48,13 @@ bool MbHasSkinColor(const uint8_t* y_src,
const int stride_v, const int stride_v,
const int mb_row, const int mb_row,
const int mb_col) { const int mb_col) {
const uint8_t* y = const uint8_t* y = y_src + ((mb_row << 4) + 8) * stride_y + (mb_col << 4) + 8;
y_src + ((mb_row << 4) + 8) * stride_y + (mb_col << 4) + 8; const uint8_t* u = u_src + ((mb_row << 3) + 4) * stride_u + (mb_col << 3) + 4;
const uint8_t* u = const uint8_t* v = v_src + ((mb_row << 3) + 4) * stride_v + (mb_col << 3) + 4;
u_src + ((mb_row << 3) + 4) * stride_u + (mb_col << 3) + 4;
const uint8_t* v =
v_src + ((mb_row << 3) + 4) * stride_v + (mb_col << 3) + 4;
// Use 2x2 average of center pixel to compute skin area. // Use 2x2 average of center pixel to compute skin area.
uint8_t y_avg = uint8_t y_avg = (*y + *(y + 1) + *(y + stride_y) + *(y + stride_y + 1)) >> 2;
(*y + *(y + 1) + *(y + stride_y) + *(y + stride_y + 1)) >> 2; uint8_t u_avg = (*u + *(u + 1) + *(u + stride_u) + *(u + stride_u + 1)) >> 2;
uint8_t u_avg = uint8_t v_avg = (*v + *(v + 1) + *(v + stride_v) + *(v + stride_v + 1)) >> 2;
(*u + *(u + 1) + *(u + stride_u) + *(u + stride_u + 1)) >> 2;
uint8_t v_avg =
(*v + *(v + 1) + *(v + stride_v) + *(v + stride_v + 1)) >> 2;
// Ignore MB with too high or low brightness. // Ignore MB with too high or low brightness.
if (y_avg < y_low || y_avg > y_high) if (y_avg < y_low || y_avg > y_high)
return false; return false;

View File

@ -43,14 +43,17 @@ void VPMVideoDecimator::SetTargetFramerate(int frame_rate) {
} }
bool VPMVideoDecimator::DropFrame() { bool VPMVideoDecimator::DropFrame() {
if (!enable_temporal_decimation_) return false; if (!enable_temporal_decimation_)
return false;
if (incoming_frame_rate_ <= 0) return false; if (incoming_frame_rate_ <= 0)
return false;
const uint32_t incomingframe_rate = const uint32_t incomingframe_rate =
static_cast<uint32_t>(incoming_frame_rate_ + 0.5f); static_cast<uint32_t>(incoming_frame_rate_ + 0.5f);
if (target_frame_rate_ == 0) return true; if (target_frame_rate_ == 0)
return true;
bool drop = false; bool drop = false;
if (incomingframe_rate > target_frame_rate_) { if (incomingframe_rate > target_frame_rate_) {
@ -91,7 +94,6 @@ bool VPMVideoDecimator::DropFrame() {
return drop; return drop;
} }
uint32_t VPMVideoDecimator::GetDecimatedFrameRate() { uint32_t VPMVideoDecimator::GetDecimatedFrameRate() {
ProcessIncomingframe_rate(TickTime::MillisecondTimestamp()); ProcessIncomingframe_rate(TickTime::MillisecondTimestamp());
if (!enable_temporal_decimation_) { if (!enable_temporal_decimation_) {

View File

@ -14,9 +14,7 @@
namespace webrtc { namespace webrtc {
VideoDenoiser::VideoDenoiser() VideoDenoiser::VideoDenoiser()
: width_(0), : width_(0), height_(0), filter_(DenoiserFilter::Create()) {}
height_(0),
filter_(DenoiserFilter::Create()) {}
void VideoDenoiser::TrailingReduction(int mb_rows, void VideoDenoiser::TrailingReduction(int mb_rows,
int mb_cols, int mb_cols,
@ -32,13 +30,13 @@ void VideoDenoiser::TrailingReduction(int mb_rows,
// do NOT denoise for the block. Set different threshold for skin MB. // do NOT denoise for the block. Set different threshold for skin MB.
// The change of denoising status will not propagate. // The change of denoising status will not propagate.
if (metrics_[mb_index].is_skin) { if (metrics_[mb_index].is_skin) {
// The threshold is high (more strict) for non-skin MB where the trailing // The threshold is high (more strict) for non-skin MB where the
// usually happen. // trailing usually happen.
if (metrics_[mb_index].denoise && if (metrics_[mb_index].denoise &&
metrics_[mb_index + 1].denoise + metrics_[mb_index + 1].denoise + metrics_[mb_index - 1].denoise +
metrics_[mb_index - 1].denoise +
metrics_[mb_index + mb_cols].denoise + metrics_[mb_index + mb_cols].denoise +
metrics_[mb_index - mb_cols].denoise <= 2) { metrics_[mb_index - mb_cols].denoise <=
2) {
metrics_[mb_index].denoise = 0; metrics_[mb_index].denoise = 0;
filter_->CopyMem16x16(mb_src, stride_y, mb_dst, stride_y); filter_->CopyMem16x16(mb_src, stride_y, mb_dst, stride_y);
} }
@ -50,7 +48,8 @@ void VideoDenoiser::TrailingReduction(int mb_rows,
metrics_[mb_index - mb_cols + 1].denoise + metrics_[mb_index - mb_cols + 1].denoise +
metrics_[mb_index - mb_cols - 1].denoise + metrics_[mb_index - mb_cols - 1].denoise +
metrics_[mb_index + mb_cols].denoise + metrics_[mb_index + mb_cols].denoise +
metrics_[mb_index - mb_cols].denoise <= 7) { metrics_[mb_index - mb_cols].denoise <=
7) {
filter_->CopyMem16x16(mb_src, stride_y, mb_dst, stride_y); filter_->CopyMem16x16(mb_src, stride_y, mb_dst, stride_y);
} }
} }
@ -91,23 +90,20 @@ void VideoDenoiser::DenoiseFrame(const VideoFrame& frame,
uint8_t y_tmp[16 * 16] = {0}; uint8_t y_tmp[16 * 16] = {0};
for (int mb_row = 0; mb_row < mb_rows; ++mb_row) { for (int mb_row = 0; mb_row < mb_rows; ++mb_row) {
for (int mb_col = 0; mb_col < mb_cols; ++mb_col) { for (int mb_col = 0; mb_col < mb_cols; ++mb_col) {
const uint8_t* mb_src = const uint8_t* mb_src = y_src + (mb_row << 4) * stride_y + (mb_col << 4);
y_src + (mb_row << 4) * stride_y + (mb_col << 4);
uint8_t* mb_dst = y_dst + (mb_row << 4) * stride_y + (mb_col << 4); uint8_t* mb_dst = y_dst + (mb_row << 4) * stride_y + (mb_col << 4);
int mb_index = mb_row * mb_cols + mb_col; int mb_index = mb_row * mb_cols + mb_col;
// Denoise each MB at the very start and save the result to a temporary // Denoise each MB at the very start and save the result to a temporary
// buffer. // buffer.
if (filter_->MbDenoise( if (filter_->MbDenoise(mb_dst, stride_y, y_tmp, 16, mb_src, stride_y, 0,
mb_dst, stride_y, y_tmp, 16, mb_src, stride_y, 0, 1) == 1) == FILTER_BLOCK) {
FILTER_BLOCK) {
uint32_t thr_var = 0; uint32_t thr_var = 0;
// Save var and sad to the buffer. // Save var and sad to the buffer.
metrics_[mb_index].var = filter_->Variance16x8( metrics_[mb_index].var = filter_->Variance16x8(
mb_dst, stride_y, y_tmp, 16, &metrics_[mb_index].sad); mb_dst, stride_y, y_tmp, 16, &metrics_[mb_index].sad);
// Get skin map. // Get skin map.
metrics_[mb_index].is_skin = metrics_[mb_index].is_skin = MbHasSkinColor(
MbHasSkinColor(y_src, u_src, v_src, stride_y, stride_u, stride_v, y_src, u_src, v_src, stride_y, stride_u, stride_v, mb_row, mb_col);
mb_row, mb_col);
// Variance threshold for skin/non-skin MB is different. // Variance threshold for skin/non-skin MB is different.
// Skin MB use a small threshold to reduce blockiness. // Skin MB use a small threshold to reduce blockiness.
thr_var = metrics_[mb_index].is_skin ? 128 : 12 * 128; thr_var = metrics_[mb_index].is_skin ? 128 : 12 * 128;

View File

@ -22,8 +22,11 @@ class VideoDenoiser {
void DenoiseFrame(const VideoFrame& frame, VideoFrame* denoised_frame); void DenoiseFrame(const VideoFrame& frame, VideoFrame* denoised_frame);
private: private:
void TrailingReduction(int mb_rows, int mb_cols, const uint8_t* y_src, void TrailingReduction(int mb_rows,
int stride_y, uint8_t* y_dst); int mb_cols,
const uint8_t* y_src,
int stride_y,
uint8_t* y_dst);
int width_; int width_;
int height_; int height_;
rtc::scoped_ptr<DenoiseMetrics[]> metrics_; rtc::scoped_ptr<DenoiseMetrics[]> metrics_;

View File

@ -16,7 +16,6 @@
#include "webrtc/base/logging.h" #include "webrtc/base/logging.h"
#include "webrtc/system_wrappers/include/critical_section_wrapper.h" #include "webrtc/system_wrappers/include/critical_section_wrapper.h"
namespace webrtc { namespace webrtc {
namespace { namespace {
@ -62,8 +61,8 @@ void VideoProcessing::GetFrameStats(const VideoFrame& frame,
} }
} }
stats->num_pixels = (width * height) / stats->num_pixels = (width * height) / ((1 << stats->sub_sampling_factor) *
((1 << stats->sub_sampling_factor) * (1 << stats->sub_sampling_factor)); (1 << stats->sub_sampling_factor));
assert(stats->num_pixels > 0); assert(stats->num_pixels > 0);
// Compute mean value of frame // Compute mean value of frame
@ -112,22 +111,19 @@ int32_t VideoProcessingImpl::Deflickering(VideoFrame* frame,
return deflickering_.ProcessFrame(frame, stats); return deflickering_.ProcessFrame(frame, stats);
} }
int32_t VideoProcessingImpl::BrightnessDetection( int32_t VideoProcessingImpl::BrightnessDetection(const VideoFrame& frame,
const VideoFrame& frame,
const FrameStats& stats) { const FrameStats& stats) {
rtc::CritScope mutex(&mutex_); rtc::CritScope mutex(&mutex_);
return brightness_detection_.ProcessFrame(frame, stats); return brightness_detection_.ProcessFrame(frame, stats);
} }
void VideoProcessingImpl::EnableTemporalDecimation(bool enable) { void VideoProcessingImpl::EnableTemporalDecimation(bool enable) {
rtc::CritScope mutex(&mutex_); rtc::CritScope mutex(&mutex_);
frame_pre_processor_.EnableTemporalDecimation(enable); frame_pre_processor_.EnableTemporalDecimation(enable);
} }
void VideoProcessingImpl::SetInputFrameResampleMode(
void VideoProcessingImpl::SetInputFrameResampleMode(VideoFrameResampling VideoFrameResampling resampling_mode) {
resampling_mode) {
rtc::CritScope cs(&mutex_); rtc::CritScope cs(&mutex_);
frame_pre_processor_.SetInputFrameResampleMode(resampling_mode); frame_pre_processor_.SetInputFrameResampleMode(resampling_mode);
} }