Backbone of MobilenetV2 with Pytorch

This is backbone use pretraining model deep learning with network mobilenetV2 and evaluate model

Dev

• Dao Duy Ngu
• Le Van Thien

Install

Anaconda

• conda create --name deep python=3.8
• conda activate deep

Packages

• pip install -r requirements.txt

Construct dataset

Example: Classification dog and cat

• if dataset not split
  • dataset:
    • dog
    • cat
  • python split_dataset.py --path-dataset PathDataset --path-save PathSave
• else:
  • dataset:
    • train
      • dog
      • cat
    • val
      • dog
      • cat
    • test
      • dog
      • cat

Training

• change numbers classes with variable num_classes in file train at line 18
  • from models.MobilenetV2 import mobilenet_v2
  • model = mobilenet_v2(pretrained=True, num_classes=2).to(device)
• run training model
  • python train.py --dataset PathDataset --epochs NumbersEpochs --batch-size SizeOfBatch --image-size SizeInput

Test

• python test.py --file-folder FolderContainImage --folder-model FolderContainModel

Evaluate model

• construct test folder example:
  • test
    • dog
    • cat
• python evaluate_model.py --folder-test FolderTest --folder-model FolderContainModel --path-save FolderSave

Reference

• Release of advanced design of MobilenetV2 ICCV2019
• Release of advanced pre-trained MobilenetV2 imagenet Pytoch