Open farida-bint opened 9 months ago
[...] and the data are sampled using a TasksetDataset
First of all note that there are only the classes Taskset
and TaskDataset
, with the latter being deprecated in favor of the first.
Question: Is it normal that after the set of tasks are created, the inputs images (the originals) are no longer there?
Not really, no. I just tested it in the script anil_fc100.py.
Maybe you can share your code?
Thanks a lot for your reply,
I already used the Taskset option and it didn't change the results. Below is a screenshot of a Colab file I used to check the input images after sampling some tasks from mini-imagenet train data.
Before sampling the tasks, the input images are all ok
After sampling, the images seem unavailable or maybe transformed
In general, it is much easier to read code when it is properly formatted in the text msg, take a look here: https://docs.github.com/en/get-started/writing-on-github/working-with-advanced-formatting/creating-and-highlighting-code-blocks :slightly_smiling_face:
You might be using a wrong order for defining the MetaDataset
and the Taskset
, take a look here: https://github.com/learnables/learn2learn/blob/master/examples/vision/anil_fc100.py#L80-L91
Hi,
I tested MAML and Reptile algorithms using mini-imagenet dataset, and the data are sampled using a TasksetDataset,
Question: Is it normal that after the set of tasks are created, the inputs images (the originals) are no longer there? For each batch in the training loop, I tried to plot the samples (images) but I got blank images presenting few points in it.
I wanted to visualize the results (predictions) of each algorithm in a form of : (input image, prediction, label) but got no image.