PaddleClas/docs/en/tutorials/quick_start_en.md

Trial in 30mins

Based on the flowers102 dataset, it takes only 30 mins to experience PaddleClas, include training varieties of backbone and pretrained model, SSLD distillation, and multiple data augmentation, Please refer to Installation to install at first.

Preparation

• enter insatallation dir

cd path_to_PaddleClas

• Enter dataset/flowers102, download and decompress flowers102 dataset.

cd dataset/flowers102
# If you want to download from the brower, you can copy the link, visit it
# in the browser, download and then commpress.
wget https://paddle-imagenet-models-name.bj.bcebos.com/data/flowers102.zip
unzip flowers102.zip

• Return PaddleClas dir

cd ../../

Environment

Download pretrained model

You can use the following commands to downdload the pretrained models.

mkdir pretrained
cd pretrained
wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_pretrained.pdparams
wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_ssld_pretrained.pdparams
wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV3_large_x1_0_pretrained.pdparams

cd ../

Note: If you want to download the pretrained models on Windows environment, you can copy the links to the browser and download.

Training

• All experiments are running on the NVIDIA® Tesla® V100 single card.

Train from scratch

• Train ResNet50_vd

export CUDA_VISIBLE_DEVICES=0
python3 tools/train.py -c ./configs/quick_start/ResNet50_vd.yaml

The validation Top1 Acc curve is shown below.

Finetune - ResNet50_vd pretrained model (Acc 79.12%)

• finetune ResNet50_vd_ model pretrained on the 1000-class Imagenet dataset

export CUDA_VISIBLE_DEVICES=0
python3 tools/train.py -c ./configs/quick_start/ResNet50_vd_finetune.yaml

The validation Top1 Acc curve is shown below

Compare with training from scratch, it improve by 65% to 94.02%

You can use the trained model to infer the result of image docs/images/quick_start/flowers102/image_06739.jpg. The command is as follows.

python3 tools/infer/infer.py \
    -i docs/images/quick_start/flowers102/image_06739.jpg \
    --model=ResNet50_vd \
    --pretrained_model="output/ResNet50_vd/best_model/ppcls" \
    --class_num=102

The output is as follows. Top-5 class ids and their scores are printed.

Current image file: docs/images/quick_start/flowers102/image_06739.jpg
    top1, class id: 0, probability: 0.5129
    top2, class id: 50, probability: 0.0671
    top3, class id: 18, probability: 0.0377
    top4, class id: 82, probability: 0.0238
    top5, class id: 54, probability: 0.0231

• Note: Results are different for different models, so you might get different results for the command.

SSLD finetune - ResNet50_vd_ssld pretrained model (Acc 82.39%)

Note: when finetuning model, which has been trained by SSLD, please use smaller learning rate in the middle of net.

ARCHITECTURE:
    name: 'ResNet50_vd'
    params:
        lr_mult_list: [0.5, 0.5, 0.6, 0.6, 0.8]
pretrained_model: "./pretrained/ResNet50_vd_ssld_pretrained"

Tringing script

export CUDA_VISIBLE_DEVICES=0
python3 tools/train.py -c ./configs/quick_start/ResNet50_vd_ssld_finetune.yaml

Compare with finetune on the 79.12% pretrained model, it improve by 0.9% to 95%.

More architecture - MobileNetV3

Training script

export CUDA_VISIBLE_DEVICES=0
python3 tools/train.py -c ./configs/quick_start/MobileNetV3_large_x1_0_finetune.yaml

Compare with ResNet50_vd pretrained model, it decrease by 5% to 90%. Different architecture generates different performance, actually it is a task-oriented decision to apply the best performance model, should consider the inference time, storage, heterogeneous device, etc.

RandomErasing

Data augmentation works when training data is small.

Training script

export CUDA_VISIBLE_DEVICES=0
python3 tools/train.py -c ./configs/quick_start/ResNet50_vd_ssld_random_erasing_finetune.yaml

It improves by 1.27% to 96.27%

• Save ResNet50_vd pretrained model to experience next chapter.

cp -r output/ResNet50_vd/19/  ./pretrained/flowers102_R50_vd_final/

Distillation

• Use extra_list.txt as unlabeled data, Note:
  • Samples in the extra_list.txt and val_list.txt don't have intersection
  • Because of in the source code, label information is unused, This is still unlabeled distillation
  • Teacher model use the pretrained_model trained on the flowers102 dataset, and student model use the MobileNetV3_large_x1_0 pretrained model(Acc 75.32%) trained on the ImageNet1K dataset

total_images: 7169
ARCHITECTURE:
    name: 'ResNet50_vd_distill_MobileNetV3_large_x1_0'
pretrained_model:
    - "./pretrained/flowers102_R50_vd_final/ppcls"
    - "./pretrained/MobileNetV3_large_x1_0_pretrained/”
TRAIN:
    file_list: "./dataset/flowers102/train_extra_list.txt"

Final training script

export CUDA_VISIBLE_DEVICES=0
python3 tools/train.py -c ./configs/quick_start/R50_vd_distill_MV3_large_x1_0.yaml

It significantly imporve by 6.47% to 96.47% with more unlabeled data and teacher model.

All accuracy

Configuration	Top1 Acc
ResNet50_vd.yaml	0.2735
MobileNetV3_large_x1_0_finetune.yaml	0.9000
ResNet50_vd_finetune.yaml	0.9402
ResNet50_vd_ssld_finetune.yaml	0.9500
ResNet50_vd_ssld_random_erasing_finetune.yaml	0.9627
R50_vd_distill_MV3_large_x1_0.yaml	0.9647

The whole accuracy curves are shown below

• NOTE: As flowers102 is a small dataset, validatation accuracy maybe float 1%.

• Please refer to Getting_started for more details