A study of backbone models

ILSVRC Evaluation

goal

• get severial ImageNet pre-trained weights
• evaluate top1 and top5 accuracy on ILSVRC 2012 validation set

prepare dataset

• Download full dataset from ILSVRC kaggle
• under Data/CLS-LOC contain following folders
  • train:
  • val: 50000 images
  • test: 100000 images
• after unzip, there is a csv file named LOC_val_solution

result

• model: mobilenet_v2 trained from keras application

python mnv2_evaluation.py