Features: - `clip_image` with frame indexes or range for video via FFmpeg - add h5 support for `io` - impl `Seq.length` and `Seq.size` to get length of the sequence - Now `backend` module is imported automatically
0.2.1
Features: - Remove `scipy`, `ipython`, `pl_bolts`, `pandas`, `pyav` as optional dependency for `tensorneko_util`
Fix: - Fix metrics `epoch` is not logged in Tensorboard
0.2.0
The new minor version `0.2.0` is released. In this version, the `pytorch-lightning` version is migrated to the latest `1.7.x` and Python 3.10 is supported.
Features: - Migrate to `pytorch-lightning` version `0.2.0`. - Support Python 3.10
0.1.40
Features: - add `yaml` in `io` module to read and write yaml files. - `Patching` layer for clip tensor into patches/cubes - Add non-trainable positional embedding `SinCosPositionalEmbedding` - `BiMap` as a bidirectional dictionary.
0.1.39
Features: - Implement `Singleton` decorator to implement class-like singleton object, inspired from Kotlin/Scala - Implement `frames2video` for merge frame images as a video
Fix: - the wrong default arg for video reader.
0.1.38
Features: - `eventbus` module in `tensorneko_util` to build event-based reactive programming
Fix: - FFmpeg checker fails in Linux/macOS - Constrain version for `torch` as 1.12 may have some issues - Remove `ffmpeg-python` dependency as it's not necessary