Video Description/Captioning Framework
Pele is a diverse and flexible library designed for research into sequence to sequence learning, primarily in hierarchical vision tasks such as video captioning, and video activity recognition. There are often many moving parts to building systems for these tasks, and Pele is designed to help alleviate much of the stress, time, and headache associated with video data, and sequence to sequence learning.
You need to have the following requirements to install the pele library:
- torch>=1.4 - torchvision>=0.5 - numpy>=1.18 - sentencepiece>=0.1.8 - gin-config>=0.3.0 - Click>=7.0 - ray>=0.9.0.dev0 - apex>=0.1 - tensorboardX>=2.0 # Unmarked Ray dependency - requests>=2.23.0 # Unmarked Ray dependency - pandas>=1.0.1 # Unmarked Ray dependency - tabulate>=0.8.6 # Unmarked Ray dependency
There are two of these dependencies which are tricky to install. Apex should be installed following the instructions at https://github.com/NVIDIA/apex, while Ray's development version can be installed by following the nightly build instructions at https://ray.readthedocs.io/en/latest/installation.html.
Once installing Apex and Ray, we can clone this repository here and then run the following from the repository directory:
pip install -e .
For all of the details, check out our documentation here.
Every experiment is specified by a gin-config file, outlining the data sources, the model, the losses, and the training procedure.
To get started, check out the sample experiments in
examples/. You can run any training experiment by using the command:
pele train [Experiment .gin File]
Pele is currently maintained by David Chan, and the team at the CannyLab at the University of California, Berkeley.
Pele is licensed under Apache 2.0, as found in the LICENSE file.
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size pele-0.1.1.tar.gz (20.4 kB)||File type Source||Python version None||Upload date||Hashes View|