search

CanPeng123 / FSCIL_ALICE

28 stars 6 forks source link

The testing results #1

Closed ZhimaoPeng closed 2 years ago

ZhimaoPeng commented 2 years ago

When I run the training code i.e. "main_base.py", I got the following results: acquiring class-wise feature prototype from training data ... 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 235/235 [00:02<00:00, 102.11it/s] acquiring feature prototype for testing data ... 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 60/60 [00:00<00:00, 73.33it/s] Session 0 | NCM accuracy: 0.8031666666666667 Saving the model at the last epoch. It seems normal. But when I run the testing code i.e. "main_inc_ncm.py",, I got the following results: => creating model 'resnet18' => loading checkpoint '/home/ubuntu/code/open-set/FSCIL_ALICE/exp/CIFAR100_base-data-2-fusion_cosFace_batch-128_epoch-100_lr-1e-2_milestones-60-70_seed-1_s-30_m-4e-1/1/trial1_session0_best.pth' => loaded pre-trained model '/home/ubuntu/code/open-set/FSCIL_ALICE/exp/CIFAR100_base-data-2-fusion_cosFace_batch-128_epoch-100_lr-1e-2_milestones-60-70_seed-1_s-30_m-4e-1/1/trial1_session0_best.pth' session: 0 number of images: 30000


Files already downloaded and verified
---- CIFAR100 Base Transform ---
~~~~~~~~ testing dataset ~~~~~~~~
---- CIFAR100 Testing Transform ---
length of the trainset: 30000
----------------------------- calculate and store average class-wise feature embedding -----------------------------
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 235/235 [00:02<00:00, 93.17it/s]
----------------------------- do interference -----------------------------
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 60/60 [00:00<00:00, 79.58it/s]
TEST, total average accuracy=0.1545
TEST, task-wise correct prediction: [927, 0, 0, 0, 0, 0, 0, 0, 0]
TEST, task-wise number of images: [6000, 0, 0, 0, 0, 0, 0, 0, 0]
TEST, task-wise accuracy: [0.1545, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
This looks weird. Did I do something wrong?  Can you provide a testing configuration? thank you!
CanPeng123 commented 2 years ago

The testing configurations have been uploaded.