-
Dear authors:
Thank you for opensourcing your great work RCG.
However, I have noticed that:
1. A closely related technique called self-conditioning (NO clustering is performed, in contrast to 3, 34,…
-
We recently updated our Elastic cluster from `7.17` to `8.15.1` and we also moved from Curator `5.8.4` to `8.0.16`.
After some run, we spotted some issues that we didn't detect before.
Some indices…
-
Каждый уважающий себя рисерчер должен в какой-то момент придумать свою собственную сетку. Вот и у меня есть такое желание. я много насмотрелся и думаю что могу сделать лучше объеденив разные идеи под …
-
Hello, when training rcg, i think training is divided into two parts:
1. Training of RDM: freeze the SSL encoder, and then calculate the MSE loss between the rep after the encoder and the rep after…
-
### 🚀 The feature, motivation and pitch
I need to observe the strength of attention for each element in the sequence with all the elements in the same sequence. This is called the attention map in …
-
There are a batch of models that fail the accuracy test in inference when freezing is enabled.
- TB: mnasnet1_0 , moco, shufflenet_v2_x1_0
- TIMM: adv_inception_v3, botnet26t_256, gluon_inception_v3…
-
### Summary
I noticed that your SSL models use weight decay for all their parameters, see for example:
https://github.com/microsoft/torchgeo/blob/0f8b0ac3eabf1e4156e7259d34b1b63e6df5d445/torchge…
-
When training RDM, I have no idea if the model converges, so could you please provide some log for reference?
-
Dear authors,
Thanks for open-sourcing your great work!!!
However, I have a question about the SSL encoder:
I have noticed MoCo-v3 models are trained on ImageNet. Would they be influenced by the…
-
Sik-Ho Tang. [Review — MoCo v2: Improved Baselines with Momentum Contrastive Learning](https://sh-tsang.medium.com/review-moco-v2-improved-baselines-with-momentum-contrastive-learning-8b03598a0324).