https://github.com/ammar257ammar/keras-inceptionv4

Keras Implementation of Google's Inception-V4 Architecture (includes Keras compatible pre-trained weights)

https://github.com/ammar257ammar/keras-inceptionv4

Science Score: 10.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • CITATION.cff file
  • codemeta.json file
  • .zenodo.json file
  • DOI references
  • Academic publication links
    Links to: arxiv.org
  • Academic email domains
  • Institutional organization owner
  • JOSS paper metadata
  • Scientific vocabulary similarity
    Low similarity (7.1%) to scientific vocabulary
Last synced: 9 months ago · JSON representation

Repository

Keras Implementation of Google's Inception-V4 Architecture (includes Keras compatible pre-trained weights)

Basic Info
  • Host: GitHub
  • Owner: ammar257ammar
  • License: apache-2.0
  • Default Branch: master
  • Homepage:
  • Size: 1.63 MB
Statistics
  • Stars: 0
  • Watchers: 0
  • Forks: 0
  • Open Issues: 0
  • Releases: 0
Fork of kentsommer/keras-inceptionV4
Created almost 5 years ago · Last pushed almost 9 years ago

https://github.com/ammar257ammar/keras-inceptionV4/blob/master/

# Keras Inception-V4
Keras implementation of Google's inception v4 model with ported weights!

As described in:
[Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning (Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, Alex Alemi)](https://arxiv.org/abs/1602.07261)

Note this Keras implementation tries to follow the [tf.slim definition](https://github.com/tensorflow/models/blob/master/slim/nets/inception_v4.py) as closely as possible.

Pre-Trained weights for this Keras model can be found here (ported from the tf.slim ckpt): https://github.com/kentsommer/keras-inceptionV4/releases

You can evaluate a sample image by performing the following (weights are downloaded automatically):
* ```$ python evaluate_image.py```
```
Loaded Model Weights!
Class is: African elephant, Loxodonta africana
Certainty is: 0.868498
```

# News
5/23/2017:

* Enabled support for both Theano and Tensorflow (again... :neckbeard:)
* Added useful training parameters 
  * l2 regularization added to conv layers
  * Variance Scaling initialization added to conv layers
  * Momentum value updated for batch_norm layers
* Updated pre-processing to match paper (subtracts 0.5 instead of 1.0 :fire:)
* Minor code changes and cleanup is also included in the recent changes




# Performance Metrics (@Top5, @Top1)

Error rate on non-blacklisted subset of ILSVRC2012 Validation Dataset (Single Crop):
* Top@1 Error: 19.54%
* Top@5 Error: 4.88%

These error rates are actually slightly lower than the listed error rates in the paper:
* Top@1 Error: 20.0%
* Top@5 Error: 5.0%

Owner

  • Name: Ammar Ammar
  • Login: ammar257ammar
  • Kind: user
  • Location: The Netherlands
  • Company: Maastricht University

GitHub Events

Total
Last Year