https://github.com/ram81/open-images-challenge-2019
https://github.com/ram81/open-images-challenge-2019
Last synced: 4 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/ram81/open-images-challenge-2019
- Owner: Ram81
- Created: 2019-10-01T05:18:10.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2019-10-01T06:28:01.000Z (over 5 years ago)
- Last Synced: 2025-01-07T19:21:30.773Z (5 months ago)
- Language: Python
- Size: 14.6 KB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
## Keras-RetinaNet for Open Images Challenge 2019
**This code was taken from the 15th place in Kaggle Google AI Open Images - Object Detection Track 2018 competition:***
https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018
Repository contains the following:
* Pre-trained models (with ResNet101 and ResNet152 backbones)
* Example code to get predictions with these models for any set of images
* Code to train your own classifier based on Keras-RetinaNet and OID dataset
* Code to expand predictions for full 500 classes
## Online demo
http://nn-box.com/box/ - upload image wait several seconds and it will show boxes. ResNet152 is used as backbone.
## Requirements
Python 3.5, Keras 2.2, [Keras-RetinaNet 0.4.1](https://github.com/fizyr/keras-retinanet)
## Pretrained models
There are 3 RetinaNet models based on ResNet50, ResNet101 and ResNet152 for [443 classes](https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/blob/master/a00_utils_and_constants.py#L36) (only Level 1).
| Backbone | Image Size (px) | Model (training) | Model (inference) | Small validation mAP | Full validation mAP |
| --- | --- | --- | --- | --- | --- |
| ResNet50 | 768 - 1024 | [533 MB](https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/releases/download/v1.1/retinanet_resnet50_level_1.h5) | [178 MB](https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/releases/download/v1.1/retinanet_resnet50_level_1_converted.h5) | 0.4621 | 0.3520 |
| ResNet101 | 768 - 1024 | [739 MB](https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/releases/download/v1.2/retinanet_resnet101_level_1_v1.2.h5) | [247 MB](https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/releases/download/v1.2/retinanet_resnet101_level_1_v1.2_converted.h5) | 0.5031 | 0.3870 |
| ResNet152 | 600 - 800 | [918 MB](https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/releases/download/v1.2/retinanet_resnet152_level_1_v1.2.h5) | [308 MB](https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/releases/download/v1.2/retinanet_resnet152_level_1_v1.2_converted.h5) | 0.5194 | 0.3959 |
* Model (training) - can be used to resume training or can be used as pretrain for your own classifier
* Model (inference) - can be used to get prediction boxes for arbitrary images
## Inference
Example can be found here: [retinanet_inference_example.py](https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/blob/master/retinanet_inference_example.py)
You need to change [files_to_process = glob.glob(DATASET_PATH + 'validation_big/\*.jpg')](https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/blob/master/retinanet_inference_example.py#L181) to your own set of files.
On output you will get "predictions_\*.csv" file with boxes.
Having these predictions you can expand it to all 500 classes using code from [create_higher_level_predictions_from_level_1_predictions_csv.py](https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/blob/master/create_higher_level_predictions_from_level_1_predictions_csv.py)
## Training
For training you need to download OID dataset (~500 GB images): https://storage.googleapis.com/openimages/web/challenge.html
Next fix paths in [a00_utils_and_constants.py](https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/blob/master/a00_utils_and_constants.py)
Then to train on OID dataset you need to run python files in following order:
* create_files_for_training_by_levels.py
* retinanet_training_level_1/find_image_parameters.py
then
* retinanet_training_level_1/train_oid_level_1_resnet101.py
or
* retinanet_training_level_1/train_oid_level_1_resnet152.py
## Ensembles
If you have predictions from several models, for example for ResNet101 and ResNet152 backbones, then you can ensemble boxes with script:
* [ensemble_predictions_with_weighted_method.py](https://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018/blob/master/ensemble_predictions_with_weighted_method.py)
Proposed method increases the overall performance:
* ResNet101 mAP 0.3776 + ResNet152 mAP 0.3840 gives in result: mAP 0.4220
## Method description
* https://www.kaggle.com/c/google-ai-open-images-object-detection-track/discussion/64633