https://github.com/kentaroy47/kaggle-nflimpact-13thplace

Solution of team tara: Public 7th, Private 13th (The renewed pipeline scores 8th place)
https://github.com/kentaroy47/kaggle-nflimpact-13thplace

3d-classification kaggle pytorch video-recognition

Last synced: about 1 month ago
JSON representation

Solution of team tara: Public 7th, Private 13th (The renewed pipeline scores 8th place)

• Host: GitHub
• URL: https://github.com/kentaroy47/kaggle-nflimpact-13thplace
• Owner: kentaroy47
• License: mit
• Created: 2021-01-26T04:42:57.000Z (almost 5 years ago)
• Default Branch: main
• Last Pushed: 2021-01-28T04:05:37.000Z (almost 5 years ago)
• Last Synced: 2025-04-03T18:09:01.532Z (8 months ago)
• Topics: 3d-classification, kaggle, pytorch, video-recognition
• Language: Jupyter Notebook
• Homepage:
• Size: 15.5 MB
• Stars: 9
• Watchers: 1
• Forks: 0
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# 13th place solution of team tara

![](https://storage.googleapis.com/kaggle-media/competitions/nfl-helmet-impacts/impacts.gif)

This is our team's solution for [Kaggle NFL 1st and Future - Impact Detection](https://www.kaggle.com/c/nfl-impact-detection)

Work by team tara: [@tereka](https://www.kaggle.com/tereka) [@hidehisaarai1213](https://www.kaggle.com/hidehisaarai1213) [@rishigami](https://www.kaggle.com/rishigami) [@arutema47](https://www.kaggle.com/kyoshioka47)

Public: 7th 0.5503

Private: 13th 0.5017

fix.. Private: 8th 0.5337

## Overview

[Solution writeup](https://www.kaggle.com/c/nfl-impact-detection/discussion/208801)

[Submitted inference code](https://www.kaggle.com/kyoshioka47/pp-mod-of-2-stage-3d?scriptVersionId=51447826)

![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3508221%2Fa5abcefa4a1968535ed22d82b18569ee%2Fimage%20(1).png?generation=1609822336590711&alt=media)

# Preparation

`pip install -r requirements.txt`

1. `train-prepare-labels.ipynb`でラベルと画像データを書き出し

Write out training images with `train-prepare-labels.ipynb`

2.`prepare_classification_images.ipynb`でclassification用データを書き出し

Write out classification images with `prepare_classification_images.ipynb`

3. pretrainフォルダにeffdetの事前学習モデルを入れておく

Place effdet pretrained models inside `pretrain` folder.

# train

for end model:

`python train_1ststage.py --enum 15 --modeltype Endzone --cutmix --strech 3 --imsize 1024 --bs 3 --all --effdet effdet4 --lr 1e-4`

for side model:

`python train_1ststage.py --enum 15 --modeltype Sideline --cutmix --strech 3 --imsize 1024 --bs 3 --all --effdet effdet4 --lr 1e-4`

classification

`python train_2ndstage.py --all`

# Check CV

1. 1ststage-inferenceでdetection結果を取得

Get inference results with `1st-stage-Inference.ipynb`

2.2nd stage-inferenceでclassification。

CLassify with `2nd-stage-Inference.ipynb`

CV: `0.47~0.5`