-
Hi, Thanks for this interesting work. I have a question regarding to barlow twins. Did you use barlow twins on the equivariance branch as well or did you keep the ntxent loss?
Also is equivariant pr…
-
**Describe the bug**
Might not be a bug but your guidance is valued:
I tried to reproduce the imagenet100 results using Barlow Twins and get accuracies well below the officially reported figures.
…
-
### Describe the bug
Same error has occurred while using Neptune's Transformers integration during fine-tuning model with Transformers' Trainer.
### Reproduction
I followed instructions available …
-
Hi vturrisi,
First thanks for your effort for this amazing repo of ssl.
In some of methods, such as barlowtwins, byol, I find two loss during pretrain stage:
https://github.com/vturrisi/solo-le…
-
It seems like there is no actual checkpoint available to download for Barlow Twins pretrained for 100 epochs on ImageNet (the drive folder only contains the args.json folder).
Is it possible to add…
-
Hi! I'm Marcus Ng, and we have corresponded previously. You mentioned that the original dataset of ~400K cif files and mofids used to joint pre-train the models together are around 4GB and uploading i…
-
Just running the example verbatim from the notebook unsupervised_hello_world.ipynb for SimClr. The train loss seemed to get stuck around 6.9287. The solution has no evidence of a collapse with proj_st…
-
I tried to run your code to reproduce Table7. But I got this result.
`run_all.sh`:
```bash
#!/bin/bash
arch=$1
mem_size=$2
dataset=$3
ntask=$4
offline_ep=$5
if [ -z "$arch" ] | [ -z "$mem…
-
# Add default parameters for all projection heads
It's helpful to know what the default parameters were in the papers to get started. We should add the default projection head parameters which were…
-
Tech Insights. [Graph contrastive learning](https://medium.com/@TechInsight/graph-contrastive-learning-b82dbef44f23).
Zhu Y, Xu Y, Liu Q, et al. [An empirical study of graph contrastive learning](htt…