Action Tubelet Detection in AVA

In this project we have adapted the Action Tubelet Detector [1], presented at ICCV17 for the Atomic Visual Action dataset from Google [2].

The code is a modified version of [https://github.com/vkalogeiton/caffe/tree/act-detector] with the changes required to adapt the code for a new dataset.

To use this code, follow the instructions from the github of the authors of the ACT and generate the metadata needed with the AVA dataset using generate_ACT_metadata_AVA.py

[1] V. Kalogeiton, et al. Action Tubelet Detector for Spatio-Temporal Action Localization at ICCV2017 [https://arxiv.org/abs/1705.01861]

[2] C. Gu, et al. AVA: A Video Dataset of Spatio-temporally Localized Atomic Visual Actions at arXiv [https://arxiv.org/abs/1705.08421]