effnet_L_finetuned / README.md
eingrid's picture
Update README.md
36546cd verified
metadata
datasets:
  - frgfm/imagenette
metrics:
  - accuracy
pipeline_tag: image-classification

Model Card for Fine-tuned EfficientNet-L on Imagenette

Model Details

Model Name: Fine-tuned EfficientNet-L

Type of Model: Image Classification

Version: 1.0

Model Training

Data Description: Trained on the Imagenette dataset, a public subset of ImageNet featuring ten easily classifiable classes from diverse categories.

Training Procedure: The model utilized pre-trained weights from EfficientNet-L and was fine-tuned across all layers. Training involved 30 planned epochs with a cosine learning rate schedule starting from 0.0005 to 0. However, training was early stopped after 10 epochs due to convergence.

Training Configuration:
num_epochs: 30.0
init_lr: 0.0005

training_modifiers:
  - !EpochRangeModifier
    start_epoch: 0.0
    end_epoch: eval(num_epochs)

  - !LearningRateFunctionModifier
    final_lr: 0.0
    init_lr: eval(init_lr)
    lr_func: cosine
    start_epoch: 0.0
    end_epoch: eval(num_epochs)

Evaluation Metrics: The model's performance was monitored using the following metrics:

Val Loss: 0.32155620951985703

Top 1 Accuracy: 92%

Usage :

Clone repo, then you can use the model in the following way

import torch
from  torchvision.models import efficientnet_v2_l

NUM_LABELS = 10
checkpoint = torch.load("/effnet_L_finetuned/pytorch_model.bin")
model = efficientnet_v2_l()
model.classifier[1] = torch.nn.Linear(model.classifier[1].in_features, NUM_LABELS)
model.load_state_dict(checkpoint['state_dict'])