rusttorch-codec 0.4.0

Typed codec pipelines for RustTorch
Documentation
<!-- Generated by scripts/check-compatibility.py. Do not edit directly. -->
# rusttorch-codec compatibility

This page is generated from [`compat/pytorch_api.toml`](../../compat/pytorch_api.toml) against PyTorch `v2.13.0` at commit `cf30153`, through `tch` `0.26.0`.
Each entry is independently scoped. Supported applies only to its written scope; row counts are not a compatibility score.

## Supported

No entries.

## Partial

### `codec`

- **PyTorch:** `torch.utils.data.Dataset for encoded media`
- **RustTorch:** `rusttorch::vision::decode_image`, `rusttorch::audio::decode_audio`, `rusttorch::codec::MediaDecoder`, `rusttorch::codec::MediaFrame`
- **Implementation:** RustTorch frontend backed by LibTorch
- **Pinned inventory:** 0 identities ([exact mapping](../../compat/pytorch_inventory_map.toml)); disposition only, not a support claim.
- **Scope:** Optional domain packages decode bounded local PNG/JPEG images, WAV/PCM and FLAC audio, and selected FFmpeg CPU audio/video streams into owned typed samples for the shared Dataset and fallible-iterator batching interfaces.
- **Pinned source:** [`torch/utils/data/dataset.py`](https://github.com/pytorch/pytorch/blob/cf30153/torch/utils/data/dataset.py)
- **Evidence:** [`crates/rusttorch-vision/tests/pipelines.rs::imagefolder_workers_decode_classify_and_bound_payload`](../../crates/rusttorch-vision/tests/pipelines.rs), [`crates/rusttorch-audio/tests/pipelines.rs::wav_decode_workers_and_padding_preserve_rates_and_lengths`](../../crates/rusttorch-audio/tests/pipelines.rs), [`crates/rusttorch-audio/tests/pipelines.rs::flac_decodes_the_same_samples_as_pcm_wav`](../../crates/rusttorch-audio/tests/pipelines.rs), [`crates/rusttorch-codec/tests/pipelines.rs::cpu_video_streams_batches_and_seeks_with_timestamps`](../../crates/rusttorch-codec/tests/pipelines.rs), [`crates/rusttorch-codec/tests/pipelines.rs::native_audio_copies_checked_float_buffers`](../../crates/rusttorch-codec/tests/pipelines.rs)
- **Notes:** Umbrella decoding integration row; vision, audio and video specify the exact format, transform and native-backend contracts. Core/data remain format-agnostic and decoding dependencies are opt-in. No universal codec registry, implicit downloads, hardware decoding or full external media-library parity. Native FFmpeg availability and licensing follow the explicitly selected system/vcpkg build.

### `video`

- **PyTorch:** `torch.utils.data.Dataset for video samples`
- **RustTorch:** `rusttorch::codec::MediaDecoder`, `rusttorch::codec::MediaFrame`, `rusttorch::codec::VideoFrame`, `rusttorch::codec::AudioFrame`, `rusttorch::codec::TimeBase`, `rusttorch::codec::Timestamp`, `rusttorch::codec::capabilities`
- **Implementation:** RustTorch frontend backed by LibTorch
- **Pinned inventory:** 0 identities ([exact mapping](../../compat/pytorch_inventory_map.toml)); disposition only, not a support claim.
- **Scope:** Optional rusttorch-codec FFmpeg 8 CPU decoding of one selected video or mono/stereo audio stream from regular local files; owned RGB CHW frames and Float audio blocks, rational presentation time, checked seeking with pre-target frame removal, lazy typed batches, cancellation checks between native operations and finite input/output/duration limits.
- **Pinned source:** [`torch/utils/data/dataset.py`](https://github.com/pytorch/pytorch/blob/cf30153/torch/utils/data/dataset.py)
- **Evidence:** [`crates/rusttorch-codec/tests/pipelines.rs::rational_time_is_checked_and_exact`](../../crates/rusttorch-codec/tests/pipelines.rs), [`crates/rusttorch-codec/tests/pipelines.rs::cpu_video_streams_batches_and_seeks_with_timestamps`](../../crates/rusttorch-codec/tests/pipelines.rs), [`crates/rusttorch-codec/tests/pipelines.rs::native_audio_copies_checked_float_buffers`](../../crates/rusttorch-codec/tests/pipelines.rs), [`crates/rusttorch-codec/tests/pipelines.rs::limits_and_missing_streams_fail_explicitly`](../../crates/rusttorch-codec/tests/pipelines.rs)
- **Notes:** Choose explicit system or vcpkg linking; no bundled native binaries. Checked fixtures use FFV1 Matroska and PCM WAV. Allowed demuxers include Matroska/WebM, WAV, AVI, MOV/MP4, MP3 and FLAC, with actual codec availability reported from the selected native build. Network/playlist input, hardware decoding, arbitrary multichannel layouts, cross-stream synchronization and exact decoder checkpointing are outside scope. The same four tests also pass against the checksummed FFmpeg 8.1.2 dynamically linked LGPL-only profile from scripts/build-ffmpeg.py; this is Linux/macOS profile support, not Windows native runtime evidence.

## Planned

No entries.

## Python-only

No entries.

## Not supported

No entries.