Skip to content
Code supporting the paper: "TenniSet: A Dataset for Dense Fine-Grained Event Recognition, Localisation and Description"
Python
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
annotator
data
models
utils
LICENSE
README.md
dataset.py
metrics.py
model.py
train.py

README.md

Tennis (Work in progress)

Code supporting the paper: "TenniSet: A Dataset for Dense Fine-Grained Event Recognition, Localisation and Description"

Code is being converted into MXNet and GluonCV. With newly trained models available soon.

PLEASE NOTE: The results in the paper were with outdated Keras models, new results will be presented below in this readme.

About

The Annotator

The annotator can be used to annotate any video with dense temporal events using a GUI. See the README in the annotator directory for more information.

Data Pre-processing

See data for download and organisation information.

Once you have .json annotation files with the annotator, you can run:

python utils/annotations/preprocess.py

This does pre-processing on the annotations, specifically:

  1. Generates slice .txt files for each .json annotation file
  2. Generalises the .json annotation files from player names and forehand/backhand to near/far and left/right
  3. Generates label .txt files for each generalised .json annotation file

Alternatively you can download our annotations .tar.gz (see data)

The Models

Coming Soon - More information on the models can be found in the README in the models directory.

Evalutations

...

Visualisations

...

You can’t perform that action at this time.