Merge pull request #644 from bot66/fix_typo

Fix typo

Author: KevinMusgrave

CONTENTS.md

@@ -21,12 +21,14 @@
 | [**IntraPairVarianceLoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#intrapairvarianceloss) | [Deep Metric Learning with Tuplet Margin Loss](http://openaccess.thecvf.com/content_ICCV_2019/papers/Yu_Deep_Metric_Learning_With_Tuplet_Margin_Loss_ICCV_2019_paper.pdf)
 | [**LargeMarginSoftmaxLoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#largemarginsoftmaxloss) | [Large-Margin Softmax Loss for Convolutional Neural Networks](https://arxiv.org/pdf/1612.02295.pdf)
 | [**LiftedStructreLoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#liftedstructureloss) | [Deep Metric Learning via Lifted Structured Feature Embedding](https://arxiv.org/pdf/1511.06452.pdf)
+| [**ManifoldLoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#manifoldloss) | [Ensemble Deep Manifold Similarity Learning using Hard Proxies](https://openaccess.thecvf.com/content_CVPR_2019/papers/Aziere_Ensemble_Deep_Manifold_Similarity_Learning_Using_Hard_Proxies_CVPR_2019_paper.pdf)
 | [**MarginLoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#marginloss) | [Sampling Matters in Deep Embedding Learning](https://arxiv.org/pdf/1706.07567.pdf)
 | [**MultiSimilarityLoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#multisimilarityloss) | [Multi-Similarity Loss with General Pair Weighting for Deep Metric Learning](http://openaccess.thecvf.com/content_CVPR_2019/papers/Wang_Multi-Similarity_Loss_With_General_Pair_Weighting_for_Deep_Metric_Learning_CVPR_2019_paper.pdf)
 | [**NCALoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#ncaloss) | [Neighbourhood Components Analysis](https://www.cs.toronto.edu/~hinton/absps/nca.pdf)
 | [**NormalizedSoftmaxLoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#normalizedsoftmaxloss) | - [NormFace: L2 Hypersphere Embedding for Face Verification](https://arxiv.org/pdf/1704.06369.pdf) <br/> - [Classification is a Strong Baseline for DeepMetric Learning](https://arxiv.org/pdf/1811.12649.pdf)
 | [**NPairsLoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#npairsloss) | [Improved Deep Metric Learning with Multi-class N-pair Loss Objective](http://www.nec-labs.com/uploads/images/Department-Images/MediaAnalytics/papers/nips16_npairmetriclearning.pdf)
 | [**NTXentLoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#ntxentloss) | - [Representation Learning with Contrastive Predictive Coding](https://arxiv.org/pdf/1807.03748.pdf) <br/> - [Momentum Contrast for Unsupervised Visual Representation Learning](https://arxiv.org/pdf/1911.05722.pdf) <br/> - [A Simple Framework for Contrastive Learning of Visual Representations](https://arxiv.org/abs/2002.05709)
+| [**P2SGradLoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#p2sgradloss) | [P2SGrad: Refined Gradients for Optimizing Deep Face Models](https://arxiv.org/abs/1905.02479)
 | [**PNPLoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#pnploss) | [Rethinking the Optimization of Average Precision: Only Penalizing Negative Instances before Positive Ones is Enough](https://arxiv.org/pdf/2102.04640.pdf)
 | [**ProxyAnchorLoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#proxyanchorloss) | [Proxy Anchor Loss for Deep Metric Learning](https://arxiv.org/pdf/2003.13911.pdf)
 | [**ProxyNCALoss**](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#proxyncaloss) | [No Fuss Distance Metric Learning using Proxies](https://arxiv.org/pdf/1703.07464.pdf)

README.md

@@ -18,13 +18,15 @@
 
 ## News
 
+**June 18**: v2.2.0
+- Added [ManifoldLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#manifoldloss) and [P2SGradLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#p2sgradloss).
+- Added a `symmetric` flag to [SelfSupervisedLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#selfsupervisedloss).
+- See the [release notes](https://github.com/KevinMusgrave/pytorch-metric-learning/releases/tag/v2.2.0).
+- Thank you [domenicoMuscill0](https://github.com/domenicoMuscill0).
+
 **April 5**: v2.1.0
 - Added [PNPLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#pnploss)
-- Thanks to contributor [interestingzhuo](https://github.com/interestingzhuo).
-
-**January 29**: v2.0.0
-- Added SelfSupervisedLoss, plus various API improvements. See the [release notes](https://github.com/KevinMusgrave/pytorch-metric-learning/releases/tag/v2.0.0).
-- Thanks to contributor [cwkeam](https://github.com/cwkeam).
+- Thanks you [interestingzhuo](https://github.com/interestingzhuo).
 
 
 ## Documentation
@@ -225,6 +227,7 @@ Thanks to the contributors who made pull requests!
 
 | Contributor | Highlights |
 | -- | -- |
+|[domenicoMuscill0](https://github.com/domenicoMuscill0)| - [ManifoldLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#manifoldloss) <br/> - [P2SGradLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#p2sgradloss) 
 |[mlopezantequera](https://github.com/mlopezantequera) | - Made the [testers](https://kevinmusgrave.github.io/pytorch-metric-learning/testers) work on any combination of query and reference sets <br/> - Made [AccuracyCalculator](https://kevinmusgrave.github.io/pytorch-metric-learning/accuracy_calculation/) work with arbitrary label comparisons |
 |[cwkeam](https://github.com/cwkeam) | - [SelfSupervisedLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#selfsupervisedloss) <br/> - [VICRegLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#vicregloss) <br/> - Added mean reciprocal rank accuracy to [AccuracyCalculator](https://kevinmusgrave.github.io/pytorch-metric-learning/accuracy_calculation/) <br/> - BaseLossWrapper|
 |[marijnl](https://github.com/marijnl)| - [BatchEasyHardMiner](https://kevinmusgrave.github.io/pytorch-metric-learning/miners/#batcheasyhardminer) <br/> - [TwoStreamMetricLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/trainers/#twostreammetricloss) <br/> - [GlobalTwoStreamEmbeddingSpaceTester](https://kevinmusgrave.github.io/pytorch-metric-learning/testers/#globaltwostreamembeddingspacetester) <br/> - [Example using trainers.TwoStreamMetricLoss](https://github.com/KevinMusgrave/pytorch-metric-learning/blob/master/examples/notebooks/TwoStreamMetricLoss.ipynb) |
@@ -273,6 +276,7 @@ This library contains code that has been adapted and modified from the following
 - https://github.com/ronekko/deep_metric_learning
 - https://github.com/tjddus9597/Proxy-Anchor-CVPR2020
 - http://kaizhao.net/regularface
+- https://github.com/nii-yamagishilab/project-NN-Pytorch-scripts
 
 ### Logo
 Thanks to [Jeff Musgrave](https://www.designgenius.ca/) for designing the logo.

docs/losses.md

@@ -545,6 +545,57 @@ losses.LiftedStructureLoss(neg_margin=1, pos_margin=0, **kwargs):
 * **loss**: The loss per positive pair in the batch. Reduction type is ```"pos_pair"```.
 
 
+## ManifoldLoss
+
+[Ensemble Deep Manifold Similarity Learning using Hard Proxies](https://openaccess.thecvf.com/content_CVPR_2019/papers/Aziere_Ensemble_Deep_Manifold_Similarity_Learning_Using_Hard_Proxies_CVPR_2019_paper.pdf)
+
+```python
+losses.ManifoldLoss(
+        l: int,
+        K: int = 50,
+        lambdaC: float = 1.0,
+        alpha: float = 0.8,
+        margin: float = 5e-4,
+        **kwargs
+    )
+```
+
+**Parameters**
+
+- **l**: embedding size.
+
+- **K**: number of proxies.
+
+- **lambdaC**: regularization weight. Used in the formula `loss = intrinsic_loss + lambdaC*context_loss`.
+    If `lambdaC=0`, then it uses only the intrinsic loss. If `lambdaC=np.inf`, then it uses only the context loss.
+
+- **alpha**: parameter of the Random Walk. Must be in the range `(0,1)`. It specifies the amount of similarity between neighboring nodes.
+
+- **margin**: margin used in the calculation of the loss.
+
+
+Example usage:
+```python
+loss_fn = ManifoldLoss(128)
+
+# use random cluster centers
+loss = loss_fn(embeddings)
+# or specify indices of embeddings to use as cluster centers
+loss = loss_fn(embeddings, indices_tuple=indices)
+```
+
+**Important notes**
+
+`labels`, `ref_emb`, and `ref_labels` are not supported for this loss function.
+
+In addition, `indices_tuple` is **not** for the output of miners. Instead, it is for a list of indices of embeddings to be used as cluster centers.
+
+
+**Default reducer**: 
+
+ - This loss returns an **already reduced** loss.
+
+
 ## MarginLoss
 [Sampling Matters in Deep Embedding Learning](https://arxiv.org/pdf/1706.07567.pdf){target=_blank}
 ```python
@@ -761,6 +812,37 @@ losses.NTXentLoss(temperature=0.07, **kwargs)
 * **loss**: The loss per positive pair in the batch. Reduction type is ```"pos_pair"```.
 
 
+
+## P2SGradLoss
+[P2SGrad: Refined Gradients for Optimizing Deep Face Models](https://arxiv.org/abs/1905.02479)
+```python
+losses.P2SGradLoss(descriptors_dim, num_classes, **kwargs)
+```
+
+**Parameters**
+
+- **descriptors_dim**: The embedding size.
+
+- **num_classes**: The number of classes in your training dataset.
+
+
+Example usage:
+```python
+loss_fn = P2SGradLoss(128, 10)
+loss = loss_fn(embeddings, labels)
+```
+
+**Important notes**
+
+`indices_tuple`, `ref_emb`, and `ref_labels` are not supported for this loss function.
+
+
+**Default reducer**: 
+
+ - This loss returns an **already reduced** loss.
+
+
+
 ## PNPLoss
 [Rethinking the Optimization of Average Precision: Only Penalizing Negative Instances before Positive Ones is Enough](https://arxiv.org/pdf/2102.04640.pdf){target=_blank}
 ```python
@@ -849,14 +931,31 @@ loss_optimizer.step()
 
 ## SelfSupervisedLoss
 
-A common use case is to have `embeddings` and `ref_emb` be augmented versions of each other. For most losses, you have to create labels to indicate which `embeddings` correspond with which `ref_emb`. `SelfSupervisedLoss` automates this.
+A common use case is to have `embeddings` and `ref_emb` be augmented versions of each other. For most losses, you have to create labels to indicate which `embeddings` correspond with which `ref_emb`. 
+
+`SelfSupervisedLoss` is a wrapper that takes care of this by creating labels internally. It assumes that:
+
+- `ref_emb[i]` is an augmented version of `embeddings[i]`.
+- `ref_emb[i]` is the only augmented version of `embeddings[i]` in the batch.
 
 ```python
+losses.SelfSupervisedLoss(loss, symmetric=True, **kwargs)
+```
+
+**Parameters**:
+
+* **loss**: The loss function to be wrapped.
+* **symmetric**: If `True`, then the embeddings in both `embeddings` and `ref_emb` are used as anchors. If `False`, then only the embeddings in `embeddings` are used as anchors.
+
+Example usage:
+
+```
 loss_fn = losses.TripletMarginLoss()
 loss_fn = SelfSupervisedLoss(loss_fn)
 loss = loss_fn(embeddings, ref_emb)
 ```
 
+
 ??? "Supported Loss Functions"
     - [AngularLoss](losses.md#angularloss)
     - [CircleLoss](losses.md#circleloss)

src/pytorch_metric_learning/utils/accuracy_calculator.py

@@ -440,7 +440,7 @@ def get_accuracy(
         ):
             raise ValueError(
                 "When ref_includes_query is True, the first len(query) elements of reference must be equal to query.\n"
-                "Likewise, the first len(query_labels) elements of reference_lbels must be equal to query_labels.\n"
+                "Likewise, the first len(query_labels) elements of reference_labels must be equal to query_labels.\n"
             )
 
         self.curr_function_dict = self.get_function_dict(include, exclude)