WEEK 4 PRESENTATION NGOC TA AIDEAN SHARGHI

SST: Single-Stream Temporal Action Proposals Single pass Input: video sentence Output: number of temporal intervals that contain an action Dataset: THUMOS 14 (20min long videos) Performs better at higher tIoU regime Handle very long testing sequences

Dense-Captioning Events in Videos Localize temporal proposal of interest Describe with natural language Dataset: ActivityNet Captions Single pass Using more strides improves recall across all values of IoU’s

Bidirectional Attentive Fusion with Context Gating for Dense Video Captioning Learn the details Modify parameters

Output sentences: a man is seen speaking to the camera and leads into a person holding a ball and a man is standing in a large group of people a man is seen speaking to the camera and leads into a man holding a stick and a man is standing in a black METEOR: 5.2925

Extract features from C3D Model axon-research/c3d-keras facebook/C3D -caffe hx173149/C3D-tensorflow

THANK YOU