Training using CUB Dataset and CIFAR10 not Learning

[MagicHealer]

Hi, I've been trying to train using two different datasets as different task. In this experiment I tried to use CUB dataset as the initial task and Cifar10 as second task. However, upon training, it is not learning. Can anyone point out what i have been doing wrong? Thank you.

Here is my source code:

!pip install avalanche-lib==0.2.1
from torch.optim import SGD
from torch.nn import CrossEntropyLoss
from avalanche.benchmarks.datasets import MNIST, CIFAR10, CUB200
from avalanche.evaluation.metrics import forgetting_metrics, accuracy_metrics,
loss_metrics, timing_metrics, cpu_usage_metrics, confusion_matrix_metrics, disk_usage_metrics
from avalanche.models import SimpleMLP, SimpleCNN
from avalanche.logging import InteractiveLogger, TextLogger, TensorboardLogger, WandBLogger
from avalanche.training.plugins import EvaluationPlugin
from avalanche.training.supervised import Naive, LwF
from avalanche.benchmarks.generators import nc_benchmark, ni_benchmark

from torchvision.transforms import Compose, ToTensor, Normalize, RandomCrop, Resize
train_transform = Compose([
Resize((224,224)),
#lambda x: x.convert("RGB"),
#RandomCrop(28, padding=4),
ToTensor(),
Normalize((0.1307,), (0.3081,))
])

test_transform = Compose([
Resize((224,224)),
lambda x: x.convert("RGB"),
#RandomCrop(28, padding=4),
ToTensor(),
Normalize((0.1307,), (0.3081,))
])

cub_train = CUB200(
'./data/mnist', train=True, download=True, transform=train_transform
)
cub_test = CUB200(
'./data/mnist', train=False, download=True, transform=test_transform
)

train_transform = Compose([
Resize((224,224)),
#RandomCrop(28, padding=4),
ToTensor(),
Normalize((0.1307,), (0.3081,))
])

test_transform = Compose([
Resize((224,224)),
#RandomCrop(28, padding=4),
ToTensor(),
Normalize((0.1307,), (0.3081,))
])
train_cifar10 = CIFAR10(
'./data/cifar10', train=True, download=True, transform= train_transform
)
test_cifar10 = CIFAR10(
'./data/cifar10', train=False, download=True, transform=test_transform
)

from avalanche.benchmarks.generators import filelist_benchmark, dataset_benchmark,
tensors_benchmark, paths_benchmark

generic_scenario = dataset_benchmark(
[cub_train, train_cifar10],
[cub_test, test_cifar10]
)

from avalanche.benchmarks.utils import AvalancheDataset
train_cub_task0 = AvalancheDataset(cub_train, task_labels=0)
test_cub_task0 = AvalancheDataset(cub_test, task_labels=0)

train_cifar10_task1 = AvalancheDataset(train_cifar10, task_labels=1)
test_cifar10_task1 = AvalancheDataset(test_cifar10, task_labels=1)

scenario_custom_task_labels = dataset_benchmark(
[train_cub_task0, train_cifar10_task1],
[test_cub_task0, test_cifar10_task1]
)

print('Without custom task labels:',
generic_scenario.train_stream[1].task_label)

print('With custom task labels:',
scenario_custom_task_labels.train_stream[1].task_label)

import torch
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(f"[INFO]: Computation device: {device}")

import torchvision
model=torchvision.models.alexnet(num_classes=200)

tb_logger = TensorboardLogger()
text_logger = TextLogger(open('log.txt', 'a'))
interactive_logger = InteractiveLogger()

eval_plugin = EvaluationPlugin(
accuracy_metrics(minibatch=True, epoch=True, experience=True, stream=True),
loss_metrics(minibatch=True, epoch=True, experience=True, stream=True),
timing_metrics(epoch=True, epoch_running=True),
forgetting_metrics(experience=True, stream=True),
# cpu_usage_metrics(experience=True),
# disk_usage_metrics(minibatch=True, epoch=True, experience=True, stream=True),
loggers=[interactive_logger, text_logger, tb_logger]#,wandb_logger]
)

cl_strategy = LwF(
model, SGD(model.parameters(), lr=0.001, momentum=0.9),
CrossEntropyLoss(), device=device, train_mb_size=128, train_epochs=20, eval_mb_size=128,
evaluator=eval_plugin, alpha=0.1, temperature=2)

print("Starting experiment...")
results = []
for experience in scenario_custom_task_labels.train_stream:
print("Start of Training for Task ", experience.current_experience)
cl_strategy.train(experience)
print("Training completed")
print("Computing accuracy on the whole test set")
results.append(cl_strategy.eval(scenario_custom_task_labels.test_stream))

and here is a sample result of training: