fine-grained-car-classification
Method | Accuracy | Note |
---|---|---|
ResNet101 Baseline | 0.9240144260664096 | input_size=(448, 448), optim=SGD, lr=0.1, epochs=30, lr_decay_rate=0.1, lr_decay_step=10(by epochs), batch_size=32 |
ResNet50+ArcFace+FocalLoss | 0.9368237781370476 | input_size=(448, 448), optim=SGD, lr=0.1, epochs=20, lr_decay_rate=0.1, lr_decay_step=10(by epochs), batch_size=64, s=16.0, m=0.50, easy_margin=False, gamma=2, alpha=1 |
ResNet50+MoCo(RandomColorJitter+RandomGrayscale+RandomHorizontalFlip) | 0.8977739087178211 | input_size=(448, 448), optim=SGD, lr=0.1, epochs=100, lr_decay_rate=0.1, lr_decay_step=10(by epochs), batch_size=32 |
ResNet50+TripletMarginLoss | 0.8865812709861958 | input_size=(448, 448), optim=SGD, lr=0.1, epochs=100, lr_decay_rate=0.1, lr_decay_step=10(by epochs), batch_size=32, margin=5.0, lambda_triplet=0.1 |
DenseNet201+AutoAugment+ArcFace+MoCo | 0.9477676905857481 | TBD |
DenseNet201+AutoAugment+ArcFace | 0.9446586245491855 | TBD |
DenseNet201+AutoAugment | 0.9235169755005597 | TBD |
DenseNet201 | 0.9109563487128467 | TBD |