Skip to content

Update C++ metadata names to match python #707

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Dan-Flores merged 14 commits into from
Jun 5, 2025
Merged

Update C++ metadata names to match python #707

Show file tree
Hide file tree
Changes from 9 commits
Commits
Show all changes
14 commits
Select commit Hold shift + click to select a range
a60968c
Update C++ metadata names to match python
May 29, 2025
6a9eca1
Update C++ metadata names to match python
May 29, 2025
df6072d
resolve merge conflicts
Dan-Flores May 30, 2025
1a8b3f8
fixes to pass metadata tests
Dan-Flores May 30, 2025
4219df4
Update minPtsSecondsFromScan, maxPtsSecondsFromScan, numFramesFromScan
Dan-Flores May 30, 2025
80ae01d
Fix name endStreamSecondsFromContent
Dan-Flores May 30, 2025
18ad24f
Update comment to reflect change
Dan-Flores May 30, 2025
0e85a5f
remove test artifact
Dan-Flores May 30, 2025
4327916
Update fields in c++ tests
Dan-Flores May 30, 2025
553e74a
Add pytest tmp_path (#706)
Dan-Flores Jun 2, 2025
223b1d0
Update minPtsFromScan and maxPtsFromScan
Dan-Flores Jun 2, 2025
8136466
Rename begin/endStreamSecondsFromContent to begin/endStreamPtsSeconds…
Dan-Flores Jun 2, 2025
314c3cd
Merge branch 'main' into align_metadata_names
Dan-Flores Jun 4, 2025
3966b94
correct merge changes
Dan-Flores Jun 4, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 8 additions & 8 deletions src/torchcodec/_core/Metadata.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,22 +25,22 @@ struct StreamMetadata {
AVMediaType mediaType;
std::optional<AVCodecID> codecId;
std::optional<std::string> codecName;
std::optional<double> durationSeconds;
std::optional<double> beginStreamFromHeader;
std::optional<int64_t> numFrames;
std::optional<double> durationSecondsFromHeader;
std::optional<double> beginStreamSecondsFromHeader;
std::optional<int64_t> numFramesFromHeader;
std::optional<int64_t> numKeyFrames;
std::optional<double> averageFps;
std::optional<double> averageFpsFromHeader;
std::optional<double> bitRate;

// More accurate duration, obtained by scanning the file.
// These presentation timestamps are in time base.
std::optional<int64_t> minPtsFromScan;
std::optional<int64_t> maxPtsFromScan;
// These presentation timestamps are in seconds.
std::optional<double> minPtsSecondsFromScan;
std::optional<double> maxPtsSecondsFromScan;
std::optional<double> beginStreamSecondsFromContent;
std::optional<double> endStreamSecondsFromContent;
// This can be useful for index-based seeking.
std::optional<int64_t> numFramesFromScan;
std::optional<int64_t> numFramesFromContent;

// Video-only fields derived from the AVCodecContext.
std::optional<int64_t> width;
Expand All @@ -58,7 +58,7 @@ struct ContainerMetadata {
int numVideoStreams = 0;
// Note that this is the container-level duration, which is usually the max
// of all stream durations available in the container.
std::optional<double> durationSeconds;
std::optional<double> durationSecondsFromHeader;
// Total BitRate level information at the container level in bit/s
std::optional<double> bitRate;
// If set, this is the index to the default audio stream.
Expand Down
49 changes: 25 additions & 24 deletions src/torchcodec/_core/SingleStreamDecoder.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -125,22 +125,22 @@ void SingleStreamDecoder::initializeDecoder() {

int64_t frameCount = avStream->nb_frames;
if (frameCount > 0) {
streamMetadata.numFrames = frameCount;
streamMetadata.numFramesFromHeader = frameCount;
}

if (avStream->duration > 0 && avStream->time_base.den > 0) {
streamMetadata.durationSeconds =
streamMetadata.durationSecondsFromHeader =
av_q2d(avStream->time_base) * avStream->duration;
}
if (avStream->start_time != AV_NOPTS_VALUE) {
streamMetadata.beginStreamFromHeader =
streamMetadata.beginStreamSecondsFromHeader =
av_q2d(avStream->time_base) * avStream->start_time;
}

if (avStream->codecpar->codec_type == AVMEDIA_TYPE_VIDEO) {
double fps = av_q2d(avStream->r_frame_rate);
if (fps > 0) {
streamMetadata.averageFps = fps;
streamMetadata.averageFpsFromHeader = fps;
}
containerMetadata_.numVideoStreams++;
} else if (avStream->codecpar->codec_type == AVMEDIA_TYPE_AUDIO) {
Expand All @@ -163,7 +163,7 @@ void SingleStreamDecoder::initializeDecoder() {

if (formatContext_->duration > 0) {
AVRational defaultTimeBase{1, AV_TIME_BASE};
containerMetadata_.durationSeconds =
containerMetadata_.durationSecondsFromHeader =
ptsToSeconds(formatContext_->duration, defaultTimeBase);
}

Expand Down Expand Up @@ -241,8 +241,8 @@ void SingleStreamDecoder::scanFileAndUpdateMetadataAndIndex() {
streamMetadata.maxPtsFromScan = std::max(
streamMetadata.maxPtsFromScan.value_or(INT64_MIN),
getPtsOrDts(packet) + packet->duration);
streamMetadata.numFramesFromScan =
streamMetadata.numFramesFromScan.value_or(0) + 1;
streamMetadata.numFramesFromContent =
streamMetadata.numFramesFromContent.value_or(0) + 1;

// Note that we set the other value in this struct, nextPts, only after
// we have scanned all packets and sorted by pts.
Expand All @@ -262,15 +262,15 @@ void SingleStreamDecoder::scanFileAndUpdateMetadataAndIndex() {
auto& streamMetadata = containerMetadata_.allStreamMetadata[streamIndex];
auto avStream = formatContext_->streams[streamIndex];

streamMetadata.numFramesFromScan =
streamMetadata.numFramesFromContent =
streamInfos_[streamIndex].allFrames.size();

if (streamMetadata.minPtsFromScan.has_value()) {
streamMetadata.minPtsSecondsFromScan =
streamMetadata.beginStreamSecondsFromContent =
*streamMetadata.minPtsFromScan * av_q2d(avStream->time_base);
}
if (streamMetadata.maxPtsFromScan.has_value()) {
streamMetadata.maxPtsSecondsFromScan =
streamMetadata.endStreamSecondsFromContent =
*streamMetadata.maxPtsFromScan * av_q2d(avStream->time_base);
}
}
Expand Down Expand Up @@ -445,7 +445,7 @@ void SingleStreamDecoder::addVideoStream(
containerMetadata_.allStreamMetadata[activeStreamIndex_];

if (seekMode_ == SeekMode::approximate &&
!streamMetadata.averageFps.has_value()) {
!streamMetadata.averageFpsFromHeader.has_value()) {
throw std::runtime_error(
"Seek mode is approximate, but stream " +
std::to_string(activeStreamIndex_) +
Expand Down Expand Up @@ -1397,9 +1397,9 @@ int64_t SingleStreamDecoder::secondsToIndexLowerBound(double seconds) {
auto& streamMetadata =
containerMetadata_.allStreamMetadata[activeStreamIndex_];
TORCH_CHECK(
streamMetadata.averageFps.has_value(),
streamMetadata.averageFpsFromHeader.has_value(),
"Cannot use approximate mode since we couldn't find the average fps from the metadata.");
return std::floor(seconds * streamMetadata.averageFps.value());
return std::floor(seconds * streamMetadata.averageFpsFromHeader.value());
}
default:
throw std::runtime_error("Unknown SeekMode");
Expand All @@ -1424,9 +1424,9 @@ int64_t SingleStreamDecoder::secondsToIndexUpperBound(double seconds) {
auto& streamMetadata =
containerMetadata_.allStreamMetadata[activeStreamIndex_];
TORCH_CHECK(
streamMetadata.averageFps.has_value(),
streamMetadata.averageFpsFromHeader.has_value(),
"Cannot use approximate mode since we couldn't find the average fps from the metadata.");
return std::ceil(seconds * streamMetadata.averageFps.value());
return std::ceil(seconds * streamMetadata.averageFpsFromHeader.value());
}
default:
throw std::runtime_error("Unknown SeekMode");
Expand All @@ -1442,10 +1442,11 @@ int64_t SingleStreamDecoder::getPts(int64_t frameIndex) {
auto& streamMetadata =
containerMetadata_.allStreamMetadata[activeStreamIndex_];
TORCH_CHECK(
streamMetadata.averageFps.has_value(),
streamMetadata.averageFpsFromHeader.has_value(),
"Cannot use approximate mode since we couldn't find the average fps from the metadata.");
return secondsToClosestPts(
frameIndex / streamMetadata.averageFps.value(), streamInfo.timeBase);
frameIndex / streamMetadata.averageFpsFromHeader.value(),
streamInfo.timeBase);
}
default:
throw std::runtime_error("Unknown SeekMode");
Expand All @@ -1460,12 +1461,12 @@ int64_t SingleStreamDecoder::getNumFrames(
const StreamMetadata& streamMetadata) {
switch (seekMode_) {
case SeekMode::exact:
return streamMetadata.numFramesFromScan.value();
return streamMetadata.numFramesFromContent.value();
case SeekMode::approximate: {
TORCH_CHECK(
streamMetadata.numFrames.has_value(),
streamMetadata.numFramesFromHeader.has_value(),
"Cannot use approximate mode since we couldn't find the number of frames from the metadata.");
return streamMetadata.numFrames.value();
return streamMetadata.numFramesFromHeader.value();
}
default:
throw std::runtime_error("Unknown SeekMode");
Expand All @@ -1476,7 +1477,7 @@ double SingleStreamDecoder::getMinSeconds(
const StreamMetadata& streamMetadata) {
switch (seekMode_) {
case SeekMode::exact:
return streamMetadata.minPtsSecondsFromScan.value();
return streamMetadata.beginStreamSecondsFromContent.value();
case SeekMode::approximate:
return 0;
default:
Expand All @@ -1488,12 +1489,12 @@ double SingleStreamDecoder::getMaxSeconds(
const StreamMetadata& streamMetadata) {
switch (seekMode_) {
case SeekMode::exact:
return streamMetadata.maxPtsSecondsFromScan.value();
return streamMetadata.endStreamSecondsFromContent.value();
case SeekMode::approximate: {
TORCH_CHECK(
streamMetadata.durationSeconds.has_value(),
streamMetadata.durationSecondsFromHeader.has_value(),
"Cannot use approximate mode since we couldn't find the duration from the metadata.");
return streamMetadata.durationSeconds.value();
return streamMetadata.durationSecondsFromHeader.value();
}
default:
throw std::runtime_error("Unknown SeekMode");
Expand Down
2 changes: 1 addition & 1 deletion src/torchcodec/_core/SingleStreamDecoder.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -121,7 +121,7 @@ class SingleStreamDecoder {
//
// Valid values for startSeconds and stopSeconds are:
//
// [minPtsSecondsFromScan, maxPtsSecondsFromScan)
// [beginStreamSecondsFromContent, endStreamSecondsFromContent)
FrameBatchOutput getFramesPlayedInRange(
double startSeconds,
double stopSeconds);
Expand Down
18 changes: 10 additions & 8 deletions src/torchcodec/_core/_metadata.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -225,26 +225,28 @@ def get_container_metadata(decoder: torch.Tensor) -> ContainerMetadata:
for stream_index in range(container_dict["numStreams"]):
stream_dict = json.loads(_get_stream_json_metadata(decoder, stream_index))
common_meta = dict(
duration_seconds_from_header=stream_dict.get("durationSeconds"),
duration_seconds_from_header=stream_dict.get("durationSecondsFromHeader"),
bit_rate=stream_dict.get("bitRate"),
begin_stream_seconds_from_header=stream_dict.get("beginStreamFromHeader"),
begin_stream_seconds_from_header=stream_dict.get(
"beginStreamSecondsFromHeader"
),
codec=stream_dict.get("codec"),
stream_index=stream_index,
)
if stream_dict["mediaType"] == "video":
streams_metadata.append(
VideoStreamMetadata(
begin_stream_seconds_from_content=stream_dict.get(
"minPtsSecondsFromScan"
"beginStreamSecondsFromContent"
),
end_stream_seconds_from_content=stream_dict.get(
"maxPtsSecondsFromScan"
"endStreamSecondsFromContent"
),
width=stream_dict.get("width"),
height=stream_dict.get("height"),
num_frames_from_header=stream_dict.get("numFrames"),
num_frames_from_content=stream_dict.get("numFramesFromScan"),
average_fps_from_header=stream_dict.get("averageFps"),
num_frames_from_header=stream_dict.get("numFramesFromHeader"),
num_frames_from_content=stream_dict.get("numFramesFromContent"),
average_fps_from_header=stream_dict.get("averageFpsFromHeader"),
**common_meta,
)
)
Expand All @@ -264,7 +266,7 @@ def get_container_metadata(decoder: torch.Tensor) -> ContainerMetadata:
streams_metadata.append(StreamMetadata(**common_meta))

return ContainerMetadata(
duration_seconds_from_header=container_dict.get("durationSeconds"),
duration_seconds_from_header=container_dict.get("durationSecondsFromHeader"),
bit_rate_from_header=container_dict.get("bitRate"),
best_video_stream_index=container_dict.get("bestVideoStreamIndex"),
best_audio_stream_index=container_dict.get("bestAudioStreamIndex"),
Expand Down
86 changes: 47 additions & 39 deletions src/torchcodec/_core/custom_ops.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -456,18 +456,20 @@ std::string get_json_metadata(at::Tensor& decoder) {

std::map<std::string, std::string> metadataMap;
// serialize the metadata into a string std::stringstream ss;
double durationSeconds = 0;
double durationSecondsFromHeader = 0;
if (maybeBestVideoStreamIndex.has_value() &&
videoMetadata.allStreamMetadata[*maybeBestVideoStreamIndex]
.durationSeconds.has_value()) {
durationSeconds =
.durationSecondsFromHeader.has_value()) {
durationSecondsFromHeader =
videoMetadata.allStreamMetadata[*maybeBestVideoStreamIndex]
.durationSeconds.value_or(0);
.durationSecondsFromHeader.value_or(0);
} else {
// Fallback to container-level duration if stream duration is not found.
durationSeconds = videoMetadata.durationSeconds.value_or(0);
durationSecondsFromHeader =
videoMetadata.durationSecondsFromHeader.value_or(0);
}
metadataMap["durationSeconds"] = std::to_string(durationSeconds);
metadataMap["durationSecondsFromHeader"] =
std::to_string(durationSecondsFromHeader);

if (videoMetadata.bitRate.has_value()) {
metadataMap["bitRate"] = std::to_string(videoMetadata.bitRate.value());
Expand All @@ -476,19 +478,20 @@ std::string get_json_metadata(at::Tensor& decoder) {
if (maybeBestVideoStreamIndex.has_value()) {
auto streamMetadata =
videoMetadata.allStreamMetadata[*maybeBestVideoStreamIndex];
if (streamMetadata.numFramesFromScan.has_value()) {
metadataMap["numFrames"] =
std::to_string(*streamMetadata.numFramesFromScan);
} else if (streamMetadata.numFrames.has_value()) {
metadataMap["numFrames"] = std::to_string(*streamMetadata.numFrames);
if (streamMetadata.numFramesFromContent.has_value()) {
metadataMap["numFramesFromHeader"] =
std::to_string(*streamMetadata.numFramesFromContent);
} else if (streamMetadata.numFramesFromHeader.has_value()) {
metadataMap["numFramesFromHeader"] =
std::to_string(*streamMetadata.numFramesFromHeader);
}
if (streamMetadata.minPtsSecondsFromScan.has_value()) {
metadataMap["minPtsSecondsFromScan"] =
std::to_string(*streamMetadata.minPtsSecondsFromScan);
if (streamMetadata.beginStreamSecondsFromContent.has_value()) {
metadataMap["beginStreamSecondsFromContent"] =
std::to_string(*streamMetadata.beginStreamSecondsFromContent);
}
if (streamMetadata.maxPtsSecondsFromScan.has_value()) {
metadataMap["maxPtsSecondsFromScan"] =
std::to_string(*streamMetadata.maxPtsSecondsFromScan);
if (streamMetadata.endStreamSecondsFromContent.has_value()) {
metadataMap["endStreamSecondsFromContent"] =
std::to_string(*streamMetadata.endStreamSecondsFromContent);
}
if (streamMetadata.codecName.has_value()) {
metadataMap["codec"] = quoteValue(streamMetadata.codecName.value());
Expand All @@ -499,8 +502,9 @@ std::string get_json_metadata(at::Tensor& decoder) {
if (streamMetadata.height.has_value()) {
metadataMap["height"] = std::to_string(*streamMetadata.height);
}
if (streamMetadata.averageFps.has_value()) {
metadataMap["averageFps"] = std::to_string(*streamMetadata.averageFps);
if (streamMetadata.averageFpsFromHeader.has_value()) {
metadataMap["averageFpsFromHeader"] =
std::to_string(*streamMetadata.averageFpsFromHeader);
}
}
if (videoMetadata.bestVideoStreamIndex.has_value()) {
Expand All @@ -523,8 +527,9 @@ std::string get_container_json_metadata(at::Tensor& decoder) {

std::map<std::string, std::string> map;

if (containerMetadata.durationSeconds.has_value()) {
map["durationSeconds"] = std::to_string(*containerMetadata.durationSeconds);
if (containerMetadata.durationSecondsFromHeader.has_value()) {
map["durationSecondsFromHeader"] =
std::to_string(*containerMetadata.durationSecondsFromHeader);
}

if (containerMetadata.bitRate.has_value()) {
Expand Down Expand Up @@ -562,30 +567,32 @@ std::string get_stream_json_metadata(

std::map<std::string, std::string> map;

if (streamMetadata.durationSeconds.has_value()) {
map["durationSeconds"] = std::to_string(*streamMetadata.durationSeconds);
if (streamMetadata.durationSecondsFromHeader.has_value()) {
map["durationSecondsFromHeader"] =
std::to_string(*streamMetadata.durationSecondsFromHeader);
}
if (streamMetadata.bitRate.has_value()) {
map["bitRate"] = std::to_string(*streamMetadata.bitRate);
}
if (streamMetadata.numFramesFromScan.has_value()) {
map["numFramesFromScan"] =
std::to_string(*streamMetadata.numFramesFromScan);
if (streamMetadata.numFramesFromContent.has_value()) {
map["numFramesFromContent"] =
std::to_string(*streamMetadata.numFramesFromContent);
}
if (streamMetadata.numFrames.has_value()) {
map["numFrames"] = std::to_string(*streamMetadata.numFrames);
if (streamMetadata.numFramesFromHeader.has_value()) {
map["numFramesFromHeader"] =
std::to_string(*streamMetadata.numFramesFromHeader);
}
if (streamMetadata.beginStreamFromHeader.has_value()) {
map["beginStreamFromHeader"] =
std::to_string(*streamMetadata.beginStreamFromHeader);
if (streamMetadata.beginStreamSecondsFromHeader.has_value()) {
map["beginStreamSecondsFromHeader"] =
std::to_string(*streamMetadata.beginStreamSecondsFromHeader);
}
if (streamMetadata.minPtsSecondsFromScan.has_value()) {
map["minPtsSecondsFromScan"] =
std::to_string(*streamMetadata.minPtsSecondsFromScan);
if (streamMetadata.beginStreamSecondsFromContent.has_value()) {
map["beginStreamSecondsFromContent"] =
std::to_string(*streamMetadata.beginStreamSecondsFromContent);
}
if (streamMetadata.maxPtsSecondsFromScan.has_value()) {
map["maxPtsSecondsFromScan"] =
std::to_string(*streamMetadata.maxPtsSecondsFromScan);
if (streamMetadata.endStreamSecondsFromContent.has_value()) {
map["endStreamSecondsFromContent"] =
std::to_string(*streamMetadata.endStreamSecondsFromContent);
}
if (streamMetadata.codecName.has_value()) {
map["codec"] = quoteValue(streamMetadata.codecName.value());
Expand All @@ -596,8 +603,9 @@ std::string get_stream_json_metadata(
if (streamMetadata.height.has_value()) {
map["height"] = std::to_string(*streamMetadata.height);
}
if (streamMetadata.averageFps.has_value()) {
map["averageFps"] = std::to_string(*streamMetadata.averageFps);
if (streamMetadata.averageFpsFromHeader.has_value()) {
map["averageFpsFromHeader"] =
std::to_string(*streamMetadata.averageFpsFromHeader);
}
if (streamMetadata.sampleRate.has_value()) {
map["sampleRate"] = std::to_string(*streamMetadata.sampleRate);
Expand Down
Loading
Loading