Run Remo on cloud server

Hello!

I have been able to run Remo on a cloud server using ngrok to have access from anywhere in order to let all my colleagues annotate the images that we use in our project. Now my question is: is there any way to load images, that are already in the server, into a dataset? I didn’t find the way to do this and i had to upload all the dataset from my computer.

Also, let me give you some suggestions for future versions:

  1. More documentation about how we can configure the app in a server. A better explanation about the remo.json file.
  2. Add the posibility to apply data augmentation to the datasets.

Hi @Juani , welcome to the forum!

You can load images in Remo from the server in two ways at the moment:

1- using the Python library from within the server itself. E.g. see this: https://remo.ai/docs/sdk/dataset/#add_data

2- adding a path to the local directories / files from the UI:

  • From within a dataset, click “Add data”
  • Then select the “Use local data” tab
  • Then enter the path(s) in the input box

With either option, Remo will link to your data in your server.


Thanks a lot for the suggestions!

Re documentation - great point.

We didn’t expand on it is because we are not explicitly supporting running Remo on cloud servers yet. We actually just enabled ngrok to support the Colab use case. But as you found out, it works :slight_smile:

Since you are using it, we will expand on the explanation on running on a server using ngrok and the config file in the next release - will keep you posted.

We will also be releasing a paid version with official support at some point - allowing for users authentication (so you can control who access your data) and faster loading (no need to tunnel with ngrok) + some other features.

Re data augmentation:

we actually thought about it, we decided to not do it for now because usually augmentation is done in memory and not saved to disk. Whereas Remo is meant more to deal with the core data (it links to the actual files).
Why would you want augmentation in Remo? to avoid writing the code, or to be able to visualize the data itself?
And would you be ok if Remo saved the augmented data to storage?

We were planning on enabling the creation of some datasets for self-supervised pretext tasks instead, as those are usually done on storage from what I have seen.

Thanks for the answer @andrea

I would prefer the augmentation to visualize the data and i wouldn’t have any problem if remo save the augmented data to storage, but those were just some opinion and suggestion.

I have another question about the remo working on a server…

Is there any way to config it for work with more that one person labeling at the same time? I mean, we have the app running on the server and we can label throught the browser but we can’t be more than one person using it at the same time because it doesn’t save the labels of one of the users.

Hi @Juani

That shouldn’t happen, we tested and used Remo with different users. One caveat:

if two users are annotating the same picture at the same time, they won’t see the updated annotation until they reload that picture. Data that will be saved is the annotations from the last one to finish.

Some questions:

  1. How many users are annotating at the same time?

  2. Can you describe better what you are experiencing? How often does it happen?

  3. Do you see the messages “Saving…” and then “All annotations saved” on top of the annotation tool? If you are using ngrok, this might cause some delays to save.


We are releasing the cloud version I mentioned very soon now, which should help with that.

If of interest, I would be happy to give you free access.

Essentially you’d be able to

  • use remo without ngrok (which should speed things up)
  • have private logins for your colleagues
  • host data in a private S3 (or as usual, on your own server)

Let me know!

re augmentation: thanks a lot for the feedback, collecting different point of views help of course.

If you are ok with saving the images, it’s a bit fiddly you can already do what you describe - you would need to save the images yourself when augmenting, and then add them to Remo (e.g. using the python lib - https://remo.ai/docs/tutorial_upload_annotations/).

We will definitely simplify it at some point…I personally think it’d be nice to not have to save the images, which adds time to the process and of course takes up space

Hi @andrea

You are right, it works fine. We have been re-testing it carefully and it works as you said. We got confused because we were two people labelling the same image at the same time.

It would be great if you can give me access for the cloud version!

1 Like