search

hytseng0509 / CrossDomainFewShot

Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation (ICLR 2020 spotlight)
323 stars 62 forks source link

GG! best accuracy 0.000000 in Feature encoder pre-training stage #27

Open e96031413 opened 3 years ago

e96031413 commented 3 years ago

command:

CUDA_VISIBLE_DEVICES=0 python3 train_baseline.py --method baseline++ --dataset cars --name baseline++ --train_aug

output:

Epoch 399 | Batch 75/257 | Loss 0.226424
Epoch 399 | Batch 100/257 | Loss 0.233635
Epoch 399 | Batch 125/257 | Loss 0.232594
Epoch 399 | Batch 150/257 | Loss 0.234371
Epoch 399 | Batch 175/257 | Loss 0.231336
Epoch 399 | Batch 200/257 | Loss 0.230566
Epoch 399 | Batch 225/257 | Loss 0.225214
Epoch 399 | Batch 250/257 | Loss 0.222642
GG! best accuracy 0.000000

No matter use PyTorch 1.4 or the latest version(PyTorch 1.7), I got the same problem.

Almostlover0608 commented 3 years ago

Have you solved it? I have same problem.

e96031413 commented 3 years ago

@Almostlover0608 Hi, I found that the train_baseline code did not calculate the best accuracy during training Thus, the output GG! best accuracy 0.000000 appears in the output.