Hi Contributors, I am trying to reproduce the results of SSRE and Fe

TypeError: pic should be PIL Image or ndarray. Got <class 'torch.Tensor'>,about g-u-n/pycil

Comments (2)

G-U-N commented on July 28, 2024 2

These issues are mainly due to that I specified the data preprocessing in their method file for better performance on CIFAR. For example, line 50. Deleting them should fix these issues.

from pycil.

kandeldeepak46 commented on July 28, 2024

Thank you for your reply. Commenting on those lines worked for me.

Moreover,

Are you aware of this error? Whenever I am trying to simulate 10 tasks of 20 classes on each task as specified in the following params

{
    "prefix": "train",
    "dataset": "imagenet1000",
    "memory_size": 0,
    "shuffle": true,
    "init_cls": 10,
    "increment": 20,
    "model_name": "fetril",
    "convnet_type": "resnet18",
    "device": ["2"],
    "seed": [1993],
    "init_epochs": 200,
    "init_lr" : 0.1,
    "init_weight_decay" : 5e-4,
    "epochs" : 50,
    "lr" : 0.05,
    "batch_size" : 128,
    "weight_decay" : 5e-4,
    "num_workers" : 8,
    "T" : 10
}

But Whenever I used larger initial classes as

{
    "prefix": "train",
    "dataset": "imagenet1000",
    "memory_size": 0,
    "shuffle": true,
    "init_cls": 100,
    "increment": 20,
    "model_name": "fetril",
    "convnet_type": "resnet18",
    "device": ["2"],
    "seed": [1993],
    "init_epochs": 200,
    "init_lr" : 0.1,
    "init_weight_decay" : 5e-4,
    "epochs" : 50,
    "lr" : 0.05,
    "batch_size" : 128,
    "weight_decay" : 5e-4,
    "num_workers" : 8,
    "T" : 10
}

it works. Does it need to have larger initial classes? Am I missing something here? regarding the arguments of JSON file?

Thank you

from pycil.

Recommend Projects