googlenet-v2-tf

Use Case and High-Level Description

The googlenet-v2-tf model is one of the Inception family, designed to perform image classification. Like the other Inception models, the googlenet-v2-tf model has been pretrained on the ImageNet image database. Originally redistributed as a checkpoint file, was converted to frozen graph. For details about this family of models, check out the paper, repository.

Steps to Reproduce Conversion to Frozen Graph

  1. Clone the original repository
    git clone https://github.com/tensorflow/models.git
    cd models/research/slim
  2. Checkout the commit that the conversion was tested on:
    git checkout 5d36f19
  3. Apply freeze.py.patch patch
    git apply path/to/freeze.py.patch
  4. Download the pretrained weights
  5. Install the dependencies:
    pip install tensorflow==1.14.0
  6. Run
    python3 freeze.py --ckpt path/to/inception_v2.ckpt --name inception_v2 --num_classes 1001 --output InceptionV2/Predictions/Softmax

Example

Specification

Metric Value
Type Classification
GFLOPs 4.058
MParams 11.185
Source framework TensorFlow*

Accuracy

Metric Original model Converted model
Top 1 74.09% 74.09%
Top 5 91.80% 91.80%

Performance

Input

Original model

Image, name - input, shape - 1,224,224,3, format is B,H,W,C where:

Channel order is RGB. Mean values - [127.5, 127.5, 127.5], scale value - 127.5

Converted model

Image, name - data, shape - 1,3,224,224, format is B,C,H,W where:

Channel order is BGR

Output

Original model

Object classifier according to ImageNet classes, name - InceptionV2/Predictions/Softmax, shape - 1,1001, output data format is B,C where:

Converted model

Object classifier according to ImageNet classes, name - InceptionV2/Predictions/Softmax, shape - 1,1001, output data format is B,C where:

Legal Information

The original model is distributed under the Apache License, Version 2.0. A copy of the license is provided in APACHE-2.0-TensorFlow.txt.