This portal has been discontinued and archived in 2021, and some functionality may no longer work as expected. Please explore the next generation of this technology and contact Grigori Fursin, the author and architect of this platform, for more information.
package:dataset-imagenet-preprocessed-using-opencv (v2.5.0)
Share
Creation date: 2019-07-25
Source: GitHub
cID: 1dc07ee0f4742028:4932bbdd2ac7a17b

Don't hesitate to get in touch if you encounter any issues or would like to discuss this community project!
Please report if this CK component works: 1 ▲ or fails: 0 ▲
Sign up to be notified when artifacts are shared or updated!

Description  

This meta package is our attempt to provide a unified Python API, CLI and JSON meta description for different package managers and building tools to automatically download and install different components (models, data sets, libraries, frameworks, tools) necessary to run portable program pipelines across evolving platforms. Our on-going project is to make the onboarding process as simple as possible via this platform. Please check this CK white paper and don't hesitate to contact us if you have suggestions or feedback!
  • Automation framework: CK
  • Development repository: ck-env
  • Source: GitHub
  • Tags: image-classification,dataset,imagenet,2012,val,preprocessed,using-opencv
  • Support for host OS: any
  • Support for target OS: any
  • How to get the stable version from this portal via the CK client:
    pip install cbench
    cb download package:dataset-imagenet-preprocessed-using-opencv --version=2.5.0 --all
    ck install package:dataset-imagenet-preprocessed-using-opencv
    ck help package
  • How to get the development version:
    pip install ck
    ck pull repo:ck-env
    ck install package:dataset-imagenet-preprocessed-using-opencv
    ck help package
  • Useful CK automation actions (CLI with the Python CK API and JSON IO):
    ck show env
    ck find env:{ID from above}
    ck virtual env:{ID from above}
  • CLI and Python API: module:package
  • Variations:
    • universal
    • for-mobilenet
    • for-resnet
    • unmutilated
    • audit.test03
    • crop.1000
    • crop.875
    • external_file
    • first.20
    • full
    • last.20
    • side.96
    • side.128
    • side.160
    • side.192
    • side.224
    • side.240
    • side.260
    • side.280
    • side.300
    • inter.linear
    • inter.area

Dependencies    

ReadMe  

Installation

$ ck install package --tags=dataset,imagenet,preprocessed,using-opencv

Details

Summary of preprocessing methods

The table below summarizes the available methods.

Preprocessing method OpenCV universal OpenCV for ResNet OpenCV for MobileNet
Additional tags universal for-resnet for-mobilenet
Supported models ResNet, MobileNet ResNet only MobileNet only
Supported platforms x86i x86 x86
Data format rgb8 (int8) rgbf32 (float32) rgbf32 (float32)
Data size 7.1G 29G 29G

Accuracy on the ImageNet 2012 validation set

The table below shows the accuracy on the ImageNet 2012 validation set (50,000 images) of the MLPerf Inference v0.5 image classification models measured - via TensorFlow (C++)

Model Metric OpenCV universal OpenCV for ResNet OpenCV for MobileNet
ResNet Top1 0.76442 0.76456 N/A
Top5 0.93074 0.93016 N/A
MobileNet non-quantized Top1 0.71676 N/A 0.71676
Top5 0.90118 N/A 0.90118
MobileNet quantized Top1 0.70700 N/A 0.70694
Top5 0.89594 N/A 0.89594
Model Metric OpenCV universal OpenCV for ResNet OpenCV for MobileNet
ResNet Top1 0.76442 0.76456 N/A
Top5 0.93074 0.93016 N/A
MobileNet non-quantized Top1 0.71676 N/A 0.71676
Top5 0.90118 N/A 0.90118
MobileNet quantized Top1 0.70762 N/A N/A (bug?)
Top5 0.89266 N/A N/A (bug?)
Additional notes

  • ResNet achieves 0.76450/0.93058 with TF-C++/TFLite, universal OpenCV preprocessing and the green channel mean of 116.6.

  • MobileNet quantized used to achieve 0.70776 with TFLite and universal OpenCV preprocessing with area interpolation.

Versions  

Files  

Comments  

Please log in to add your comments!
If you notice any inapropriate content that should not be here, please report us as soon as possible and we will try to remove it within 48 hours!