1 /* 2 * Copyright (c) 2015 The WebRTC project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11 #include "webrtc/modules/rtp_rtcp/source/h264_sps_parser.h" 12 13 #include "testing/gtest/include/gtest/gtest.h" 14 15 #include "webrtc/base/arraysize.h" 16 #include "webrtc/base/bitbuffer.h" 17 18 namespace webrtc { 19 20 // Example SPS can be generated with ffmpeg. Here's an example set of commands, 21 // runnable on OS X: 22 // 1) Generate a video, from the camera: 23 // ffmpeg -f avfoundation -i "0" -video_size 640x360 camera.mov 24 // 25 // 2) Scale the video to the desired size: 26 // ffmpeg -i camera.mov -vf scale=640x360 scaled.mov 27 // 28 // 3) Get just the H.264 bitstream in AnnexB: 29 // ffmpeg -i scaled.mov -vcodec copy -vbsf h264_mp4toannexb -an out.h264 30 // 31 // 4) Open out.h264 and find the SPS, generally everything between the first 32 // two start codes (0 0 0 1 or 0 0 1). The first byte should be 0x67, 33 // which should be stripped out before being passed to the parser. 34 35 static const size_t kSpsBufferMaxSize = 256; 36 37 // Generates a fake SPS with basically everything empty but the width/height. 38 // Pass in a buffer of at least kSpsBufferMaxSize. 39 // The fake SPS that this generates also always has at least one emulation byte 40 // at offset 2, since the first two bytes are always 0, and has a 0x3 as the 41 // level_idc, to make sure the parser doesn't eat all 0x3 bytes. 42 void GenerateFakeSps(uint16_t width, uint16_t height, uint8_t buffer[]) { 43 uint8_t rbsp[kSpsBufferMaxSize] = {0}; 44 rtc::BitBufferWriter writer(rbsp, kSpsBufferMaxSize); 45 // Profile byte. 46 writer.WriteUInt8(0); 47 // Constraint sets and reserved zero bits. 48 writer.WriteUInt8(0); 49 // level_idc. 50 writer.WriteUInt8(0x3u); 51 // seq_paramter_set_id. 52 writer.WriteExponentialGolomb(0); 53 // Profile is not special, so we skip all the chroma format settings. 54 55 // Now some bit magic. 56 // log2_max_frame_num_minus4: ue(v). 0 is fine. 57 writer.WriteExponentialGolomb(0); 58 // pic_order_cnt_type: ue(v). 0 is the type we want. 59 writer.WriteExponentialGolomb(0); 60 // log2_max_pic_order_cnt_lsb_minus4: ue(v). 0 is fine. 61 writer.WriteExponentialGolomb(0); 62 // max_num_ref_frames: ue(v). 0 is fine. 63 writer.WriteExponentialGolomb(0); 64 // gaps_in_frame_num_value_allowed_flag: u(1). 65 writer.WriteBits(0, 1); 66 // Next are width/height. First, calculate the mbs/map_units versions. 67 uint16_t width_in_mbs_minus1 = (width + 15) / 16 - 1; 68 69 // For the height, we're going to define frame_mbs_only_flag, so we need to 70 // divide by 2. See the parser for the full calculation. 71 uint16_t height_in_map_units_minus1 = ((height + 15) / 16 - 1) / 2; 72 // Write each as ue(v). 73 writer.WriteExponentialGolomb(width_in_mbs_minus1); 74 writer.WriteExponentialGolomb(height_in_map_units_minus1); 75 // frame_mbs_only_flag: u(1). Needs to be false. 76 writer.WriteBits(0, 1); 77 // mb_adaptive_frame_field_flag: u(1). 78 writer.WriteBits(0, 1); 79 // direct_8x8_inferene_flag: u(1). 80 writer.WriteBits(0, 1); 81 // frame_cropping_flag: u(1). 1, so we can supply crop. 82 writer.WriteBits(1, 1); 83 // Now we write the left/right/top/bottom crop. For simplicity, we'll put all 84 // the crop at the left/top. 85 // We picked a 4:2:0 format, so the crops are 1/2 the pixel crop values. 86 // Left/right. 87 writer.WriteExponentialGolomb(((16 - (width % 16)) % 16) / 2); 88 writer.WriteExponentialGolomb(0); 89 // Top/bottom. 90 writer.WriteExponentialGolomb(((16 - (height % 16)) % 16) / 2); 91 writer.WriteExponentialGolomb(0); 92 93 // Get the number of bytes written (including the last partial byte). 94 size_t byte_count, bit_offset; 95 writer.GetCurrentOffset(&byte_count, &bit_offset); 96 if (bit_offset > 0) { 97 byte_count++; 98 } 99 100 // Now, we need to write the rbsp into bytes. To do that, we'll need to add 101 // emulation 0x03 bytes if there's ever a sequence of 00 00 01 or 00 00 00 01. 102 // To be simple, just add a 0x03 after every 0x00. Extra emulation doesn't 103 // hurt. 104 for (size_t i = 0; i < byte_count;) { 105 // The -3 is intentional; we never need to write an emulation byte if the 00 106 // is at the end. 107 if (i < byte_count - 3 && rbsp[i] == 0 && rbsp[i + 1] == 0) { 108 *buffer++ = rbsp[i]; 109 *buffer++ = rbsp[i + 1]; 110 *buffer++ = 0x3u; 111 i += 2; 112 } else { 113 *buffer++ = rbsp[i]; 114 ++i; 115 } 116 } 117 } 118 119 TEST(H264SpsParserTest, TestSampleSPSHdLandscape) { 120 // SPS for a 1280x720 camera capture from ffmpeg on osx. Contains 121 // emulation bytes but no cropping. 122 const uint8_t buffer[] = {0x7A, 0x00, 0x1F, 0xBC, 0xD9, 0x40, 0x50, 0x05, 123 0xBA, 0x10, 0x00, 0x00, 0x03, 0x00, 0xC0, 0x00, 124 0x00, 0x2A, 0xE0, 0xF1, 0x83, 0x19, 0x60}; 125 H264SpsParser parser = H264SpsParser(buffer, arraysize(buffer)); 126 EXPECT_TRUE(parser.Parse()); 127 EXPECT_EQ(1280u, parser.width()); 128 EXPECT_EQ(720u, parser.height()); 129 } 130 131 TEST(H264SpsParserTest, TestSampleSPSVgaLandscape) { 132 // SPS for a 640x360 camera capture from ffmpeg on osx. Contains emulation 133 // bytes and cropping (360 isn't divisible by 16). 134 const uint8_t buffer[] = {0x7A, 0x00, 0x1E, 0xBC, 0xD9, 0x40, 0xA0, 0x2F, 135 0xF8, 0x98, 0x40, 0x00, 0x00, 0x03, 0x01, 0x80, 136 0x00, 0x00, 0x56, 0x83, 0xC5, 0x8B, 0x65, 0x80}; 137 H264SpsParser parser = H264SpsParser(buffer, arraysize(buffer)); 138 EXPECT_TRUE(parser.Parse()); 139 EXPECT_EQ(640u, parser.width()); 140 EXPECT_EQ(360u, parser.height()); 141 } 142 143 TEST(H264SpsParserTest, TestSampleSPSWeirdResolution) { 144 // SPS for a 200x400 camera capture from ffmpeg on osx. Horizontal and 145 // veritcal crop (neither dimension is divisible by 16). 146 const uint8_t buffer[] = {0x7A, 0x00, 0x0D, 0xBC, 0xD9, 0x43, 0x43, 0x3E, 147 0x5E, 0x10, 0x00, 0x00, 0x03, 0x00, 0x60, 0x00, 148 0x00, 0x15, 0xA0, 0xF1, 0x42, 0x99, 0x60}; 149 H264SpsParser parser = H264SpsParser(buffer, arraysize(buffer)); 150 EXPECT_TRUE(parser.Parse()); 151 EXPECT_EQ(200u, parser.width()); 152 EXPECT_EQ(400u, parser.height()); 153 } 154 155 TEST(H264SpsParserTest, TestSyntheticSPSQvgaLandscape) { 156 uint8_t buffer[kSpsBufferMaxSize] = {0}; 157 GenerateFakeSps(320u, 180u, buffer); 158 H264SpsParser parser = H264SpsParser(buffer, arraysize(buffer)); 159 EXPECT_TRUE(parser.Parse()); 160 EXPECT_EQ(320u, parser.width()); 161 EXPECT_EQ(180u, parser.height()); 162 } 163 164 TEST(H264SpsParserTest, TestSyntheticSPSWeirdResolution) { 165 uint8_t buffer[kSpsBufferMaxSize] = {0}; 166 GenerateFakeSps(156u, 122u, buffer); 167 H264SpsParser parser = H264SpsParser(buffer, arraysize(buffer)); 168 EXPECT_TRUE(parser.Parse()); 169 EXPECT_EQ(156u, parser.width()); 170 EXPECT_EQ(122u, parser.height()); 171 } 172 173 } // namespace webrtc 174
