Thoughts on PyTorch and Remo-Python

Hi all,

We are working on our remo-python library, and specifically on building a closer integration with different Deep Learning frameworks (starting point: PyTorch)

I’m wondering whether this is something the community appreciates. Any thought or comment on something you’d like to see is welcome!

The idea is that you can load Images and Annotations in Remo, and then directly pass them to a custom Dataset class.

Our current approach involves a light integration, using annotation files as an integration point.


  • easily visualize and browse images, predictions and annotations (directly in Jupyter Notebooks or in a separate window)

  • ability to quickly check statistics on datasets

  • flexibility in slicing data, without moving it around: you can create virtual train/test/splits or have data in different folders or even select specific images using tags

  • a more standardized interface across different tasks (e.g. ImageFolder requires a very specific folder structure)

Example video
Here we compare visualizing datasets in Jupyter: classic Python vs using Remo

Example code
In terms of code, the key is to define a custom Dataset class such that:

  • takes in an annotation file
  • loads in only the needed images based on some index (our “tags”)
  • loads the relevant image in get_item

For an Image Classification task, this looks like this:

class FlowerDataset(Dataset):
    Custom PyTorch Dataset Class to facilitate loading data for the Image Classifcation Task
    def __init__(self, annotations, train_test_valid_split, mapping = None, mode = 'train', transform = None):
            annotations: The path to the annotations CSV file. Format: file_name, class_name
            train_test_valid_split: The path to the tags CSV file for train, test, valid split. Format: file_name, tag
            mapping: a dictionary containing mapping of class name and class index. Format : {'class_name' : 'class_index'}, Default: None
            mode: Mode in which to instantiate class. Default: 'train'
            transform: The transforms to be applied to the image data

            image : Torch Tensor, label_tensor : Torch Tensor, file_name : str
        my_data = pd.read_csv(annotations, index_col='file_name')
        my_data['tag'] = pd.read_csv(train_test_valid_split, index_col='file_name')
        my_data = my_data.reset_index()

        self.mapping = mapping
        self.transform = transform
        self.mode = mode
        my_data = my_data.loc[my_data['tag'] == mode].reset_index(drop=True) = my_data
    def __len__(self):
        return len(

    def __getitem__(self, idx):
        if self.mapping is not None:
            labels = int(self.mapping[[idx, 'class_name'].lower()])
            labels = int([idx, 'class_name'])
        im_path =[idx, 'file_name']
        label_tensor =  torch.as_tensor(labels, dtype=torch.long)
        im =
        if self.transform:
            im = self.transform(im)
        if self.mode == 'test':
            # For saving the predictions, the file name is required
            return {'im' : im, 'labels': label_tensor, 'im_name' :[idx, 'file_name']}
            return {'im' : im, 'labels' : label_tensor}