Requirements Model Architecture - Ensemble about clvision-challenge-2023 HOT 4 CLOSED

Hi @Tsebeb

In general both multi-pass inference and model ensembles with the same backbone are allowed. However, the total GPU memory usage when all models are transferred to the GPU, shouldn't exceed the limit set for the challenge during training.

from clvision-challenge-2023.

@neuperc thanks a lot for the quick response.

Another quick questions regarding DataAugmentation. In the Avalanche Framework this is handled via the transform_groups in the individual streams correct? Is it allowed to add modify those or where would be the best point to add those for training?

from clvision-challenge-2023.

Another quick questions regarding DataAugmentation. In the Avalanche Framework this is handled via the transform_groups in the individual streams correct?

It's handled via transform groups in individual experiences. For this challenge, we have set default transformations for the datasets of all experiences in the stream as below:

train_transform = transforms.Compose(
        [
            transforms.RandomCrop(32, padding=4),
            transforms.RandomHorizontalFlip(),
            transforms.ToTensor(),
            transforms.Normalize(
                (0.5071, 0.4865, 0.4409), (0.2673, 0.2564, 0.2762)
            ),
        ]
    )

But you are allowed you change the transformations as you want. You can also change the data loader to adapt it to your strategy.

where would be the best point to add those for training?

For more details about dataset transformations in Avalanche you can check this link:
https://avalanche.continualai.org/how-tos/avalanchedataset/avalanche-transformations#replacing-transformations

from clvision-challenge-2023.

thank you @neuperc

from clvision-challenge-2023.