Lightning-flash

Changed

- Added GATE backbone for Tabular integrations ([1559](https://github.com/Lightning-AI/lightning-flash/pull/1559))

Fixed

- Fixed datamodule can't load files with square brackets in names ([1501](https://github.com/Lightning-AI/lightning-flash/pull/1501))
- Fixed channel dim selection on segmentation target ([1509](https://github.com/Lightning-AI/lightning-flash/pull/1509))
- Fixed used of `jsonargparse` avoiding reliance on non-public internal logic ([1620](https://github.com/Lightning-AI/lightning-flash/pull/1620))
- Compatibility with `pytorch-tabular>=1.0` ([1545](https://github.com/Lightning-AI/lightning-flash/pull/1545))
- Compatibility latest `numpy` ([1595](https://github.com/Lightning-AI/lightning-flash/pull/1595))

Added

- Added support for CLIP backbones to the `TextClassifier` and `ImageClassifier` tasks ([1458](https://github.com/Lightning-AI/lightning-flash/pull/1458))

Fixed

- Fixed compatibility with `lightning==1.8.0` ([1479](https://github.com/Lightning-AI/lightning-flash/pull/1479))
- Fixed the error message to suggest installing `icevision`, if it's not found while loading data ([1474](https://github.com/Lightning-AI/lightning-flash/pull/1474))
- Fixed compatibility with `torchmetrics==0.10.0` ([1469](https://github.com/Lightning-AI/lightning-flash/pull/1469))
- Fixed type of `n_gram` from bool to int in TranslationTask ([1486](https://github.com/Lightning-AI/lightning-flash/pull/1486))

Added

- Added support for `from_tensors` for `VideoClassification` ([1389](https://github.com/Lightning-AI/lightning-flash/pull/1389))
- Added fine-tuning strategies for DeepSpeed (with parameter loading and storing omitted) ([1377](https://github.com/Lightning-AI/lightning-flash/pull/1377))
- Added `torchvision` as a requirement to `datatype_audio.txt` as it's used for Audio Classification ([1425](https://github.com/Lightning-AI/lightning-flash/pull/1425))
- Added `figsize` and `limit_nb_samples` for showing batch images ([1381](https://github.com/Lightning-AI/lightning-flash/pull/1381))
- Added support for `from_lists` for Tabular Classification and Regression ([1337](https://github.com/PyTorchLightning/lightning-flash/pull/1337))
- Added support for `from_dicts` for Tabular Classification and Regression ([1331](https://github.com/PyTorchLightning/lightning-flash/pull/1331))
- Added support for using the `ImageEmbedder` SSL training for all image classifier backbones ([1264](https://github.com/PyTorchLightning/lightning-flash/pull/1264))
- Added support for audio file formats to `AudioClassificationData` ([1085](https://github.com/PyTorchLightning/lightning-flash/pull/1085))
- Added support for Flash serve to the `ObjectDetector` ([1370](https://github.com/PyTorchLightning/lightning-flash/pull/1370))
- Added support for loading `ImageClassificationData` from PIL images with `from_images` ([1372](https://github.com/PyTorchLightning/lightning-flash/pull/1372))
- Added support for loading `ObjectDetectionData` with `from_numpy`, `from_images`, and `from_tensors` ([1372](https://github.com/PyTorchLightning/lightning-flash/pull/1372))
- Added support for remote data loading with fsspec ([1387](https://github.com/PyTorchLightning/lightning-flash/pull/1387))
- Added support for TSV files to `from_csv` methods ([1387](https://github.com/PyTorchLightning/lightning-flash/pull/1387))
- Added support for more formats when loading audio files ([1387](https://github.com/PyTorchLightning/lightning-flash/pull/1387))
- Added support to use any task as an embedder by calling `as_embedder` ([1396](https://github.com/PyTorchLightning/lightning-flash/pull/1396))
- Added support for normalization of images in `SemanticSegmentationData` ([1399](https://github.com/PyTorchLightning/lightning-flash/pull/1399))

Changed

- Changed the `ImageEmbedder` dependency on VISSL to optional ([1276](https://github.com/PyTorchLightning/lightning-flash/pull/1276))
- Changed the transforms in `SemanticSegmentationData` to use albumentations instead of Kornia ([1313](https://github.com/PyTorchLightning/lightning-flash/pull/1313))

Removed

- Removed support for audio files with `sd2` extension, because SoundFile (for sd2 extension) doesn't accept fsspec objects ([1409](https://github.com/Lightning-AI/lightning-flash/pull/1409))

Fixed

- Fixed when suitable error not being raised for image segmentation (kornia) ([1425](https://github.com/Lightning-AI/lightning-flash/pull/1425))
- Fixed the script of integrating `lightning-flash` with `learn2learn` ([1376](https://github.com/Lightning-AI/lightning-flash/pull/1383))
- Fixed JIT tracing tests where the model class was not attached to the `Trainer` class ([1410](https://github.com/Lightning-AI/lightning-flash/pull/1410))
- Fixed examples for BaaL integration by removing usage of `on_<stage>_dataloader` hooks (removed in PL 1.7.0) ([1410](https://github.com/Lightning-AI/lightning-flash/pull/1410))
- Fixed examples for BaaL integration for the case when `probabilities` list is empty ([1410](https://github.com/Lightning-AI/lightning-flash/pull/1410))
- Fixed a bug where collate functions were not being attached successfully after the `DataLoader` is initialized (in PL 1.7.0 changing attributes after initialization doesn't do anything) ([1410](https://github.com/Lightning-AI/lightning-flash/pull/1410))
- Fixed a bug where grayscale images were not properly converted to RGB when loaded. ([1394](https://github.com/PyTorchLightning/lightning-flash/pull/1394))
- Fixed a bug where size of mask for instance segmentation doesn't match to size of original image. ([1353](https://github.com/PyTorchLightning/lightning-flash/pull/1353))
- Fixed image classification data `show_train_batch` for subplots with rows > 1. ([1339](https://github.com/PyTorchLightning/lightning-flash/pull/1315))
- Fixed support for all the versions (including the latest and older) of `baal` ([1315](https://github.com/PyTorchLightning/lightning-flash/pull/1315))
- Fixed a bug where a loaded `TabularClassifier` or `TabularRegressor` checkpoint could not be served ([1324](https://github.com/PyTorchLightning/lightning-flash/pull/1324))
- Fixed a bug where the `freeze_unfreeze` and `unfreeze_milestones` finetuning strategies could not be used in tandem with a `onecyclelr` LR scheduler ([1329](https://github.com/PyTorchLightning/lightning-flash/pull/1329))
- Fixed a bug where the backbone learning rate would be divided by 10 when unfrozen if using the `freeze_unfreeze` or `unfreeze_milestones` strategies ([1329](https://github.com/PyTorchLightning/lightning-flash/pull/1329))
- Fixed naming of optimizer and scheduler registries which did not allow manual optimization. ([1342](https://github.com/PyTorchLightning/lightning-flash/pull/1342))
- Fixed a bug where the `processor_backbone` argument to `SpeechRecognition` was not used for decoding outputs ([1362](https://github.com/PyTorchLightning/lightning-flash/pull/1362))
- Fixed a bug where `.npy` files could not be used with `SemanticSegmentationData` ([1369](https://github.com/PyTorchLightning/lightning-flash/pull/1369))

Fixed

- Fixed a bug where LR schedulers from HuggingFace could not be used with newer versions of PyTorch Lightning ([1307](https://github.com/PyTorchLightning/lightning-flash/pull/1307))
- Fixed a bug where the default Flash zero configurations for `ObjectDetector`, `InstanceSegmentation`, and `KeypointDetector` would error with the latest version of some requirements ([1306](https://github.com/PyTorchLightning/lightning-flash/pull/1306))
- Fixed plain `LightningModule` support for Flash data modules ([1281](https://github.com/PyTorchLightning/lightning-flash/pull/1281))

Fixed

- Fixed a bug where some backbones were incorrectly listed as available for the `ObjectDetector`, `InstanceSegmentation`, and `KeypointDetector` ([1267](https://github.com/PyTorchLightning/lightning-flash/pull/1267))
- Fixed a bug where the backbone would not be frozen when finetuning the `SpeechRecognition` task ([1275](https://github.com/PyTorchLightning/lightning-flash/pull/1275))
- Fixed a bug where the backbone would not be frozen when finetuning the `QuestionAnswering` task with certain model types ([1275](https://github.com/PyTorchLightning/lightning-flash/pull/1275))

Fixed

- Fixed examples (question answering), where NLTK's `punkt` module needs to be downloaded first. ([1215](https://github.com/PyTorchLightning/lightning-flash/pull/1215/files))
- Fixed normalizing inputs to video classification ([1213](https://github.com/PyTorchLightning/lightning-flash/pull/1213))
- Fixed a bug where `pretraining_transforms` in the `ImageEmbedder` was never called. ([1196](https://github.com/PyTorchLightning/lightning-flash/pull/1196))
- Fixed a bug where `BASE_MODEL_NAME` was not in the dict for dino and moco strategies. ([1196](https://github.com/PyTorchLightning/lightning-flash/pull/1196))
- Fixed support for `torch==1.11.0` ([1234](https://github.com/PyTorchLightning/lightning-flash/pull/1234))
- Fixed DDP spawn support for `ObjectDetector`, `InstanceSegmentation`, and `KeypointDetector` ([1222](https://github.com/PyTorchLightning/lightning-flash/pull/1222))
- Fixed a bug where `InstanceSegmentation` would fail if samples had an inconsistent number of bboxes, labels, and masks (these will now be treated as negative samples) ([1222](https://github.com/PyTorchLightning/lightning-flash/pull/1222))
- Fixed a bug where collate functions were never called in the `ImageEmbedder` class. ([1217](https://github.com/PyTorchLightning/lightning-flash/pull/1217))
- Fixed a bug where `ObjectDetector`, `InstanceSegmentation`, and `KeypointDetector` would log train and validation metrics with the same name ([1252](https://github.com/PyTorchLightning/lightning-flash/pull/1252))
- Fixed a bug where using `ReduceLROnPlateau` would raise an error ([1251](https://github.com/PyTorchLightning/lightning-flash/pull/1251))
- Fixed GPU support for self-supervised training with the `ImageEmbedder` ([1256](https://github.com/PyTorchLightning/lightning-flash/pull/1256))