Overview on kaggle: https://www.kaggle.com/c/dogs-vs-cats/overview
numpy pandas torch >=1.8.0 torchvision tensorboard scikit-learn
Use: python main.py
(temporarily)
Epoch | Train_Loss | Train_Acc | Valid_Loss | Valid_Acc |
---|---|---|---|---|
0 | 0.0468 | 98.457 | 0.0304 | 98.828 |
1 | 0.0126 | 99.547 | 0.0300 | 99.141 |
2 | 0.0009 | 99.701 | 0.0421 | 98.574 |
3 | 0.0069 | 99.756 | 0.0365 | 99.043 |
4 | 0.0049 | 99.851 | 0.0620 | 97.852 |
We merged resnet50, inceptionv3 and efficientnet-b4 with a concat of output feature vector. Then using a dropout layer with p=5
.
Epoch | Train_Loss | Train_Acc | Valid_Loss | Valid_Acc |
---|---|---|---|---|
0 | 0.0505 | 98.21 | 0.0278 | 99.3 |
1 | 0.0183 | 99.42 | 0.0276 | 99.2 |
2 | 0.0116 | 99.69 | 0.0299 | 98.0 |
run main.py
use function with prefix like test_merge(model_path='./tmp/merge_effi/epoch_num_1.pth',batch_size=100,csv_path="./results/final/test.csv",binary=True)
.