ContextQMD
Back to Tensorflow

Project Name

tensorflow/lite/java/ovic/Winner_OSS_Template.md

2.21.04.1 KB
Original Source
<!-- • This is a README.md template we encourage you to use when you release your model. • There are general sections we added to this template for various ML models. • You may need to add or remove sections depending on your needs. -->

Project Name

Authors

The 1st place winner of the 4th On-device Visual Intelligence Competition (OVIC) of Low-Power Computer Vision Challenge (LPCVC)

Description

<!-- Provide description of the model -->

The model submitted for the OVIC and full implementation code for training, evaluation, and inference

  • OVIC track: Image Classification, Object Detection

Algorithm

<!-- Provide details of the algorithms used -->

Requirements

<!-- • Provide description of the model • Provide brief information of the algorithms used -->

To install requirements:

setup
pip install -r requirements.txt

Pre-trained Models

ModelDownloadMD5 checksum
Model NameDownload Link (Size: KB)MD5 checksum

The model tar file contains the followings:

  • Trained model checkpoint
  • Frozen trained model
  • TensorFlow Lite model

Results

4th OVIC Public Ranked Leaderboard

Image Classification (from the Leaderboard)

RankUsernameLatencyAccuracy on Classified# ClassifiedAccuracy/TimeMetricReference Accuracy
1Usernamexx.x0.xxxx20000.0xxx0.xxxxx0.xxxxx
  • Metric: Accuracy improvement over the reference accuracy from the Pareto optimal curve
  • Accuracy on Classified: The accuracy in [0, 1] computed based only on the images classified within the wall-time
  • # Classified: The number of images classified within the wall-time
  • Accuracy/Time: The accuracy divided by either the total inference time or the wall-time, whichever is longer
  • Reference accuracy: The reference accuracy of models from the Pareto optimal curve that have the same latency as the submission

Object Detection

RankUsernameMetricRuntimemAP over timemAP of processed
1Username0.xxxxxxxx.xxxxxxx
  • Metric: COCO mAP computed on the entire minival dataset
  • mAP over time: COCO mAP on the minival dataset divided by latency per image
  • mAP of processed: COCO mAP computed only on the processed images

Dataset

<!-- • Provide detailed information of the dataset used -->

Training

<!-- • Provide detailed training information (preprocessing, hyperparameters, random seeds, and environment) • Provide a command line example for training. -->

Please run this command line for training.

shell
python3 ...

Evaluation

<!-- • Provide evaluation script with details of how to reproduce results. • Describe data preprocessing / postprocessing steps • Provide a command line example for evaluation. -->

Please run this command line for evaluation.

shell
python3 ...

References

<!-- Link to references -->

License

<!-- • Place your license text in a file named LICENSE.txt (or LICENSE.md) in the root of the repository. • Please also include information about your license in this README.md file. e.g., [Adding a license to a repository](https://help.github.com/en/github/building-a-strong-community/adding-a-license-to-a-repository) -->

This project is licensed under the terms of the Apache License 2.0.

Citation

<!-- If you want to make your repository citable, please follow the instructions at [Making Your Code Citable](https://guides.github.com/activities/citable-code/) -->

If you want to cite this repository in your research paper, please use the following information.