Torchmetrics

Added

- Added `SensitivityAtSpecificity` metric to classification subpackage ([2217](https://github.com/Lightning-AI/torchmetrics/pull/2217))
- Added `QualityWithNoReference` metric to image subpackage ([2288](https://github.com/Lightning-AI/torchmetrics/pull/2288))
- Added a new segmentation metric:
- `MeanIoU` ([1236](https://github.com/PyTorchLightning/metrics/pull/1236))
- `GeneralizedDiceScore` ([1090](https://github.com/Lightning-AI/metrics/pull/1090))
- Added support for calculating segmentation quality and recognition quality in `PanopticQuality` metric ([2381](https://github.com/Lightning-AI/torchmetrics/pull/2381))
- Added `pretty-errors` for improving error prints ([2431](https://github.com/Lightning-AI/torchmetrics/pull/2431))
- Added support for `torch.float` weighted networks for FID and KID calculations ([2483](https://github.com/Lightning-AI/torchmetrics/pull/2483))
- Added `zero_division` argument to selected classification metrics ([2198](https://github.com/Lightning-AI/torchmetrics/pull/2198))

Changed

- Made `__getattr__` and `__setattr__` of `ClasswiseWrapper` more general ([2424](https://github.com/Lightning-AI/torchmetrics/pull/2424))

Fixed

- Fix getitem for metric collection when prefix/postfix is set ([2430](https://github.com/Lightning-AI/torchmetrics/pull/2430))
- Fixed axis names with Precision-Recall curve ([2462](https://github.com/Lightning-AI/torchmetrics/pull/2462))
- Fixed list synchronization with partly empty lists ([2468](https://github.com/Lightning-AI/torchmetrics/pull/2468))
- Fixed memory leak in metrics using list states ([2492](https://github.com/Lightning-AI/torchmetrics/pull/2492))
- Fixed bug in computation of `ERGAS` metric ([2498](https://github.com/Lightning-AI/torchmetrics/pull/2498))
- Fixed `BootStrapper` wrapper not working with `kwargs` provided argument ([2503](https://github.com/Lightning-AI/torchmetrics/pull/2503))
- Fixed warnings being suppressed in `MeanAveragePrecision` when requested ([2501](https://github.com/Lightning-AI/torchmetrics/pull/2501))
- Fixed corner-case in `binary_average_precision` when only negative samples are provided ([2507](https://github.com/Lightning-AI/torchmetrics/pull/2507))

---

Fixed

- Fixed negative variance estimates in certain image metrics ([2378](https://github.com/Lightning-AI/torchmetrics/pull/2378))
- Fixed dtype being changed by deepspeed for certain regression metrics ([2379](https://github.com/Lightning-AI/torchmetrics/pull/2379))
- Fixed plotting of metric collection when prefix/postfix is set ([2429](https://github.com/Lightning-AI/torchmetrics/pull/2429))
- Fixed bug when `top_k>1` and `average="macro"` for classification metrics ([2423](https://github.com/Lightning-AI/torchmetrics/pull/2423))
- Fixed case where label prediction tensors in classification metrics were not validated correctly ([2427](https://github.com/Lightning-AI/torchmetrics/pull/2427))
- Fixed how auc scores are calculated in `PrecisionRecallCurve.plot` methods ([2437](https://github.com/Lightning-AI/torchmetrics/pull/2437))

Fixed

- Fixed how backprop is handled in `LPIPS` metric ([2326](https://github.com/Lightning-AI/torchmetrics/pull/2326))
- Fixed `MultitaskWrapper` not being able to be logged in lightning when using metric collections ([2349](https://github.com/Lightning-AI/torchmetrics/pull/2349))
- Fixed high memory consumption in `Perplexity` metric ([2346](https://github.com/Lightning-AI/torchmetrics/pull/2346))
- Fixed cached network in `FeatureShare` not being moved to the correct device ([2348](https://github.com/Lightning-AI/torchmetrics/pull/2348))
- Fix naming of statistics in `MeanAveragePrecision` with custom max det thresholds ([2367](https://github.com/Lightning-AI/torchmetrics/pull/2367))
- Fixed custom aggregation in retrieval metrics ([2364](https://github.com/Lightning-AI/torchmetrics/pull/2364))
- Fixed initialize aggregation metrics with default floating type ([2366](https://github.com/Lightning-AI/torchmetrics/pull/2366))
- Fixed plotting of confusion matrices ([2358](https://github.com/Lightning-AI/torchmetrics/pull/2358))

Added

- Added more tokenizers for `SacreBLEU` metric ([2068](https://github.com/Lightning-AI/torchmetrics/pull/2068))
- Added support for logging `MultiTaskWrapper` directly with lightnings `log_dict` method ([2213](https://github.com/Lightning-AI/torchmetrics/pull/2213))
- Added `FeatureShare` wrapper to share submodules containing feature extractors between metrics ([2120](https://github.com/Lightning-AI/torchmetrics/pull/2120))
- Added new metrics to image domain:
* `SpatialDistortionIndex` ([2260](https://github.com/Lightning-AI/torchmetrics/pull/2260))
* Added `CriticalSuccessIndex` ([2257](https://github.com/Lightning-AI/torchmetrics/pull/2257))
* `Spatial Correlation Coefficient` ([2248](https://github.com/Lightning-AI/torchmetrics/pull/2248))
- Added `average` argument to multiclass versions of `PrecisionRecallCurve` and `ROC` ([2084](https://github.com/Lightning-AI/torchmetrics/pull/2084))
- Added confidence scores when `extended_summary=True` in `MeanAveragePrecision` ([2212](https://github.com/Lightning-AI/torchmetrics/pull/2212))
- Added `RetrievalAUROC` metric ([2251](https://github.com/Lightning-AI/torchmetrics/pull/2251))
- Added `aggregate` argument to retrieval metrics ([2220](https://github.com/Lightning-AI/torchmetrics/pull/2220))
- Added utility functions in `segmentation.utils` for future segmentation metrics ([2105](https://github.com/Lightning-AI/torchmetrics/pull/2105))


Changed

- Changed minimum supported Pytorch version from 1.8 to 1.10 ([2145](https://github.com/Lightning-AI/torchmetrics/pull/2145))
- Changed x-/y-axis order for `PrecisionRecallCurve` to be consistent with scikit-learn ([2183](https://github.com/Lightning-AI/torchmetrics/pull/2183))

Deprecated

- Deprecated `metric._update_called` ([2141](https://github.com/Lightning-AI/torchmetrics/pull/2141))
- Deprecated `specicity_at_sensitivity` in favour of `specificity_at_sensitivity` ([2199](https://github.com/Lightning-AI/torchmetrics/pull/2199))

Fixed

- Fixed support for half precision + CPU in metrics requiring topk operator ([2252](https://github.com/Lightning-AI/torchmetrics/pull/2252))
- Fixed warning incorrectly being raised in `Running` metrics ([2256](https://github.com/Lightning-AI/torchmetrics/pull/2265))
- Fixed integration with custom feature extractor in `FID` metric ([2277](https://github.com/Lightning-AI/torchmetrics/pull/2277))

---

Added

- Added error if `NoTrainInceptionV3` is being initialized without `torch-fidelity` not being installed ([2143](https://github.com/Lightning-AI/torchmetrics/pull/2143))
- Added support for Pytorch v2.1 ([2142](https://github.com/Lightning-AI/torchmetrics/pull/2142))

Changed

- Change default state of `SpectralAngleMapper` and `UniversalImageQualityIndex` to be tensors ([2089](https://github.com/Lightning-AI/torchmetrics/pull/2089))
- Use `torch` range func and repeat for deterministic bincount ([2184](https://github.com/Lightning-AI/torchmetrics/pull/2184))

Removed

- Removed unused `lpips` third-party package as dependency of `LearnedPerceptualImagePatchSimilarity` metric ([2230](https://github.com/Lightning-AI/torchmetrics/pull/2230))

Fixed

- Fixed numerical stability bug in `LearnedPerceptualImagePatchSimilarity` metric ([2144](https://github.com/Lightning-AI/torchmetrics/pull/2144))
- Fixed numerical stability issue in `UniversalImageQualityIndex` metric ([2222](https://github.com/Lightning-AI/torchmetrics/pull/2222))
- Fixed incompatibility for `MeanAveragePrecision` with `pycocotools` backend when too little `max_detection_thresholds` are provided ([2219](https://github.com/Lightning-AI/torchmetrics/pull/2219))
- Fixed support for half precision in Perplexity metric ([2235](https://github.com/Lightning-AI/torchmetrics/pull/2235))
- Fixed device and dtype for `LearnedPerceptualImagePatchSimilarity` functional metric ([2234](https://github.com/Lightning-AI/torchmetrics/pull/2234))
- Fixed bug in `Metric._reduce_states(...)` when using `dist_sync_fn="cat"` ([2226](https://github.com/Lightning-AI/torchmetrics/pull/2226))
- Fixed bug in `CosineSimilarity` where 2d is expected but 1d input was given ([2241](https://github.com/Lightning-AI/torchmetrics/pull/2241))
- Fixed bug in `MetricCollection` when using compute groups and `compute` is called more than once ([2211](https://github.com/Lightning-AI/torchmetrics/pull/2211))

Added

- Added metric to cluster package:
- `MutualInformationScore` ([2008](https://github.com/Lightning-AI/torchmetrics/pull/2008))
- `RandScore` ([2025](https://github.com/Lightning-AI/torchmetrics/pull/2025))
- `NormalizedMutualInfoScore` ([2029](https://github.com/Lightning-AI/torchmetrics/pull/2029))
- `AdjustedRandScore` ([2032](https://github.com/Lightning-AI/torchmetrics/pull/2032))
- `CalinskiHarabaszScore` ([2036](https://github.com/Lightning-AI/torchmetrics/pull/2036))
- `DunnIndex` ([2049](https://github.com/Lightning-AI/torchmetrics/pull/2049))
- `HomogeneityScore` ([2053](https://github.com/Lightning-AI/torchmetrics/pull/2053))
- `CompletenessScore` ([2053](https://github.com/Lightning-AI/torchmetrics/pull/2053))
- `VMeasureScore` ([2053](https://github.com/Lightning-AI/torchmetrics/pull/2053))
- `FowlkesMallowsIndex` ([2066](https://github.com/Lightning-AI/torchmetrics/pull/2066))
- `AdjustedMutualInfoScore` ([2058](https://github.com/Lightning-AI/torchmetrics/pull/2058))
- `DaviesBouldinScore` ([2071](https://github.com/Lightning-AI/torchmetrics/pull/2071))
- Added `backend` argument to `MeanAveragePrecision` ([2034](https://github.com/Lightning-AI/torchmetrics/pull/2034))

---