guarin

Guarin lig 1572 implement msn paper

1

TODO: - [x] projection head - [x] collate function - [x] loss - [ ] examples - [ ] benchmark - [ ] docs - [ ] add to api...

SimMIM: A Simple Framework for Masked Image Modeling

5

## SimMIM: A Simple Framework for Masked Image Modeling 18.11.2021 https://arxiv.org/abs/2111.09886 https://github.com/microsoft/SimMIM Similar architecture as [MAE](https://arxiv.org/abs/2111.06377) but uses only a single linear layer as decoder instead of a transformer, passes...

## Odin: Object discovery and representation networks 16.03.2022 https://arxiv.org/abs/2203.08777 New deepmind paper about self-supervised pretraining using object segmentation. This paper builds on the [DetCon](https://arxiv.org/abs/2103.10957) paper which relies on hand-crafted segmentation...

## MoCo v3: An Empirical Study of Training Self-Supervised Vision Transformers 05.04.2021 https://arxiv.org/abs/2104.02057 https://github.com/facebookresearch/moco-v3 Adapts MoCo v2 for vision transformers and slightly simplifies the architecture (no memory queue). All the...

DirectCLR: Understanding Dimensional Collapse in Contrastive Self-supervised Learning

5

## DirectCLR: Understanding Dimensional Collapse in Contrastive Self-supervised Learning 18.10.2021 https://arxiv.org/abs/2110.09348 https://github.com/facebookresearch/directclr Similar model as SimCLR but does not require a projection head and instead calculates the loss only on...

## VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning 21.05.2021 https://arxiv.org/abs/2105.04906 https://github.com/facebookresearch/vicreg New model similar to [Barlow Twins](https://arxiv.org/abs/2103.03230) but combines three losses: invariance, variance, and covariance. Uses a fully symmetric model architecture...

## SplitMask: Are Large-scale Datasets Necessary for Self-Supervised Pre-training? 20.12.2021 https://arxiv.org/abs/2112.10740 Similar method as [MAE](https://arxiv.org/abs/2111.06377) but adds an additional InfoNCE loss and makes two instead of just one forward pass...

If the batch size in the Imagenette benchmark is chosen so that the last batch from the test dataloader has size 1 the following error is raised: ``` "Dimension out...

I tried to upload a dataset with points in its name and got the error message below. We should improve the error message to highlight that it failed due to...

Using `LightlyDataset.from_torch_dataset` with a non-classification dataset fails because lightly assumes that the target is an integer but targets can be tuples or dicts for segmentation or detection tasks. The current...