UntrimmedNet for Action Recognition and Detection

We provide the code and models for our CVPR paper (Arxiv Preprint):

  UntrimmedNets for Weakly Supervised Action Recognition and Detection
  Limin Wang, Yuanjun Xiong, Dahua Lin, and Luc Van Gool
  in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017

Updates

• September 19th, 2017

• Release the learned models on the THUMOS14 and ActivityNet1.2 datasets.

• August 20th, 2017

• Release the model protos.

Guide

The training of UntrimmedNet is composed of three steps: - Step 1: extract action proposals (or shot boundaries) for each untrimmed video. We provide a sample of detected shot boudary on the ActivityNet (v1.2) under the folders of data/anet1.2/anet_1.2_train_window_shot/ and data/anet1.2/anet1.2/anet_1.2_val_window_shot/. - Step 2: construct file lists for training and validation. There are two filelists: one containing file path, number of frames, and label; the other one containing the shot file path and number of frames (Examples are in the folder data/anet1.2/). - Step 3: train UntrimmedNets using our modified caffe: https://github.com/yjxiong/caffe/tree/untrimmednet

The testing of UntrimmedNet for action recognition is based on temporal sliding window and top-k pooling

The testing of UntrimmedNet for action detection is based on a simple baseline (see code in matlab/

Downloads

You could download our trained models on the THUMOS14 and ActivityNet datasets by using the scripts of scripts/get_reference_model_thumos.sh and scripts/get_reference_model_anet.sh.