Skip to content

Latest commit

 

History

History
84 lines (69 loc) · 3.89 KB

TRAINING.md

File metadata and controls

84 lines (69 loc) · 3.89 KB
Raw

Dependencies

Google Speech Commands

Before training, execute ./download_speech_commands_dataset.sh to download the speech commands data set.

VGG19 BN

  • accuracy: 97.337235%, 97.527432% with crop, Kaggle private LB score: 0.87454 and 0.88030 with crop, epoch time: 1m25s
python train_speech_commands.py --model=vgg19_bn --optim=sgd --lr-scheduler=plateau --learning-rate=0.01 --lr-scheduler-patience=5 --max-epochs=70 --batch-size=96

VGG19 BN with Mixup

  • accuracy: 97.483541%, 97.542063% with crop, Kaggle private LB score: 0.89521 and 0.89839 with crop, epoch time: 1m30s
python train_speech_commands.py --model=vgg19_bn --optim=sgd --lr-scheduler=plateau --learning-rate=0.01 --lr-scheduler-patience=5 --max-epochs=70 --batch-size=96 --mixup

WideResNet 28-10

  • accuracy: 97.937089%, 97.922458% with crop, Kaggle private LB score: 0.88546 and 0.88699 with crop, epoch time: 2m5s
python train_speech_commands.py --model=wideresnet28_10 --optim=sgd --lr-scheduler=plateau --learning-rate=0.01 --lr-scheduler-patience=5 --max-epochs=70 --batch-size=96

WideResNet 28-10D

  • accuracy: 97.702999%, 97.717630% with crop, Kaggle private LB score:0.89580 and 0.89568 with crop, epoch time: 2m10s
python train_speech_commands.py --model=wideresnet28_10D --optim=sgd --lr-scheduler=plateau --learning-rate=0.01 --lr-scheduler-patience=5 --max-epochs=70 --batch-size=96

WideResNet 52-10

  • accuracy: 98.039503%, 97.980980% with crop, Kaggle private LB score: 0.88159 and 0.88323 with crop, epoch time: 3m55s
python train_speech_commands.py --model=wideresnet52_10 --optim=sgd --lr-scheduler=plateau --learning-rate=0.01 --lr-scheduler-patience=5 --max-epochs=70 --batch-size=96

ResNext29 8x64

  • accuracy: 97.190929%, 97.161668% with crop, Kaggle private LB score: 0.89533 and 0.89733 with crop, epoch time: 4m36
python train_speech_commands.py --model=resnext29_8_64 --optim=sgd --lr-scheduler=plateau --learning-rate=0.01 --lr-scheduler-patience=5 --max-epochs=70 --batch-size=96

DPN92

  • accuracy: 97.190929%, 97.249451% with crop, Kaggle private LB score: 0.89075 and 0.89286 with crop, epoch time: 3m45s
python train_speech_commands.py --model=dpn92 --optim=sgd --lr-scheduler=plateau --learning-rate=0.01 --lr-scheduler-patience=5 --max-epochs=70 --batch-size=96

DenseNet-BC (L=100, k=12)

  • accuracy: 97.161668%, 97.147037% with crop, Kaggle private LB score: 0.88946 and 0.89134 with crop, epoch time: 1m30s
python train_speech_commands.py --model=densenet_bc_100_12 --optim=sgd --lr-scheduler=plateau --learning-rate=0.01 --lr-scheduler-patience=5 --max-epochs=70 --batch-size=64

DenseNet-BC (L=190, k=40)

  • accuracy: 97.117776%, 97.147037% with crop, Kaggle private LB score: 0.89369 and 0.89521 with crop, epoch time: 20m (P6000)
python train_speech_commands.py --model=densenet_bc_190_40 --optim=sgd --lr-scheduler=plateau --learning-rate=0.01 --lr-scheduler-patience=5 --max-epochs=70 --batch-size=64

CIFAR10

VGG19 BN

  • accuracy: 93.56%, epoch time: 19s
python train_cifar10.py --model=vgg19_bn --optim=sgd --learning-rate=0.1 --lr-scheduler=step --lr-scheduler-step-size=60 --max-epochs=180

WideResNet 28-10D

  • accuracy: 96.22%, epoch time: ?
python train_cifar10.py --model=wideresnet28_10D --optim=sgd --learning-rate=0.1 --lr-scheduler=step --lr-scheduler-step-size=60 --max-epochs=240 --lr-scheduler-gamma=0.2 --weight-decay=5e-4

DenseNet-BC (L=100, k=12)

  • accuracy: 95.52%, epoch time: 1m17s
python train_cifar10.py --model=densenet_bc_100_12 --optim sgd --lr-scheduler=step --learning-rate=0.1 --lr-scheduler-gamma=0.1 --lr-scheduler-step=130 --max-epochs=390 --weight-decay=1e-4 --train-batch-size=64