https://github.com/tcapelle/action_recognition

Solving UCF-101 with fastai2
https://github.com/tcapelle/action_recognition

Last synced: 7 days ago
JSON representation

Solving UCF-101 with fastai2

• Host: GitHub
• URL: https://github.com/tcapelle/action_recognition
• Owner: tcapelle
• License: apache-2.0
• Created: 2020-04-24T16:00:27.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2023-04-12T05:45:56.000Z (over 1 year ago)
• Last Synced: 2024-06-11T19:14:38.164Z (5 months ago)
• Language: Jupyter Notebook
• Homepage: https://tcapelle.github.io/action_recognition/
• Size: 17.6 MB
• Stars: 28
• Watchers: 3
• Forks: 4
• Open Issues: 5
• Metadata Files:
  • Readme: README.md
  • Contributing: CONTRIBUTING.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Action Recognition

> Implementation of various architectures to solve the UCF 101 actions dataset

It is based on the implementations found on [Action Recognition](https://github.com/eriklindernoren/Action-Recognition).

I try to keep with updated architectures that come out. Right now transformers are all we need... Follow @lucidrains to get the next attention based model ASAP.

## Install

First install `fastai`:

```bash

$ pip install fastcore fastai

```

## Results

Results are computed on a random splut 80%/20%. Using `fastai2` built-in `fit_one_cycle` training.

- [train baseline](01_train_baseline.ipynb): Implements a Basic Resnet 34 encoder coupled with a simple attention layer over the frames. (91% accuracy)

- [train convlstm](02_train_convlstm.ipynb): resnet34 encoder + LSTM layer over image features. (84.8% accuracy)

- [train_transformer](03_train_transformer.ipynb): Added the new TimeSformer and STAM from @lucidrains implementations.

This package also provides function to download nad process the video dataset into multiple frames.