-
https://arxiv.org/pdf/2108.12178.pdf
-
This issue is for discussing the idea I had in my mind (?) Basically the current architecture of slot attention can be decomposed into two parts: **core** (encoder+slot attention) and **reconstruction…
-
## 🐛 Bug
In the MoCo implementation, we need to add a head to get the MoCo v2 version of the method. However, in the momentum update, only the encoder's parameters are updated, not the head. For ex…
-
The following function apply moving average to the ema model. But it didn't update the statistic(runing_mean and runing_var) since these two were not parameters but buffers.
```
def update_moving_av…
-
Hi lucidrains,
Try this and it will NaN within 100 steps (latest Github code). The loss looks fine before NaN.
```
import torch
torch.backends.cudnn.allow_tf32 = True
torch.backends.cuda.matm…
-
### Branch
1.x branch (1.x version, such as `v1.0.0rc2`, or `dev-1.x` branch)
### Prerequisite
- [X] I have searched [Issues](https://github.com/open-mmlab/mmselfsup/issues) and [Discussions](https…
-
I use the default pretrain config to pretrain the model in ImageNet and CIFAR10,and I finetune in CIFAR10.However,it can't achieve a good result.Is there any problem in the process of my attempy.
-
# Type check lightly.models
* Work through the excluded files with prefix lightly.models and tests/models in [mypy.ini](https://github.com/lightly-ai/lightly/blob/master/mypy.ini) and type check one …
-
Dear community,
We are excited to announce the release of a new and upgraded deep learning pre-trained models library, MMPreTrain. We have integrated the original MMClassification, image classifica…
-
Hello,
I was trying to modify the code to work on the CIFAR-10 dataset and I'm having some trouble. I have changed the following:
1. LR = 0.03
2. Weight Decay = 0.0005
3. Momentum = 0.9
4. Batc…