alexsu52/mobilenet_v2_imagenette

MobileNet V2 model from Torchvision fine-tuned for Imagenette dataset. Checkpoint trained for 100 epoches using https://github.com/alexsu52/mobilenet_v2_imagenette.

Top-1 accuracy is 98.64%.

The main intent is to use it in samples and demos for model optimization. Here is the advantages:

• Imagenette can automatically downloaded and is quite small.
• MobileNet v2 is easy to train and lightweight model which is also representative and used in many public benchmarks.

Here is the code to load the checkpoint in PyTorch:

import torch
import torchvision.models as models

CHECKPOINT_URL = 'https://huggingface.co/alexsu52/mobilenet_v2_imagenette/resolve/main/pytorch_model.bin'
IMAGENETTE_CLASSES = 10

model = models.mobilenet_v2(num_classes=IMAGENETTE_CLASSES)  
        
checkpoint = torch.hub.load_state_dict_from_url(CHECKPOINT_URL, progress=False)
model.load_state_dict(checkpoint['state_dict'])