Skip to content

This is a pytorch implementation of Feedback-Network (CVPR 2017, Zamir et al.)

Notifications You must be signed in to change notification settings

LiMeng95/pytorch_feedback-network

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

PAPER

This is a pytorch implementation of Feedback-Network (CVPR 2017, Zamir et al.)

Requirements

  • Pytorch = 0.3.1
  • python = 2.7
  • numpy >= 1.14.2

Train

  • Data:

    • Download CIFAR100 dataset (binary file), and put it under folder ./data/
  • Run:

    • Run classifier_train.py
    python FeedbackNet_train.py
    • The trained model will saved in folder ./models/ every 10 epochs。
    • Attention:
      • You can adjust the parameter batch_size to fit your GPU memory。

Evaluation

  • Data:

    • Download CIFAR100 dataset (binary file), and put it under folder ./data/
  • Evaluate on CIFAR100 dataset:

    • Put the checkpoint files under folder ./models/, and specify the checkpoints file path by setting parameter params.ckpt in file classifier_train.py. You can download my baseline model here
    • Run classifier_train.py
    python FeedbackNet_test.py

Result

  • Val Accuracy
Physical / Virtual Depth PAPER(tp1) PAPER(tp5) ME(tp1) ME(tp5)
12 / 48 (stack-3;iteration=4) 71.12% 91.51% 70.92% 92.02%

To Do

  • Skip connections
  • Multi-GPU
  • Other virtual depths

Reference

About

This is a pytorch implementation of Feedback-Network (CVPR 2017, Zamir et al.)

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages