Skip to content

fuchami/scene_detection_pytorch

Repository files navigation

scene_detection_pytorch

My research repository  

Scene Detection using Multimodal Deep Triplet Networks

alpha=36のほうが良いっぽい

TODO List

  • [ ]: eval metricsの計算

  • [ ]: 各シーンのフレーム一覧

  • x: 10sec Audioでの特徴抽出

  • x: tensorboard Embedding

  • x: TripletNetのT-SNE Embedding

  • x: Compact Bilinear Poolingの実装

Memo...

  • LossのL2項
  • distanceのl2 distance
  • Online TripletNetworks

Code Structure

  • data_preprocessing.py
    • データの前処理を行う
  • eda.py
    • データのEDAを行う

Reference

・PyTorch transforms/Dataset/DataLoaderの基本動作を確認する: https://qiita.com/takurooo/items/e4c91c5d78059f92e76d

・pytorch siamese-triplet: https://github.com/adambielski/siamese-triplet

・PyTorch/TorchVisionで複数の入力をモデルに渡したいケース: https://blog.shikoan.com/torchvision-multiple-inputs/

・pytorch siamese: https://vaaaaaanquish.hatenablog.com/entry/2019/02/23/214036

・BERT Tutorial: http://mccormickml.com/2019/05/14/BERT-word-embeddings-tutorial/#3-extracting-embeddings

・VGGish: https://github.com/harritaylor/torchvggish

・Places-365 pre-trained models: https://github.com/CSAILVision/places365

https://zhuanlan.zhihu.com/p/85438252

Usage TensorBoard

ssh username@IP_Address -L 6006:localhost:6006
tensorboard --logdir <log_dir>

About

My research theme  

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published