Disentangling Affinity and Differential Embeddings
The visualizations highlight the dynamic (moving) part of the video by features of different layers (1, 2,3, and 5 from left to right) of 3D-ResNet for SSV2, HMDB51, and UCF101. The red part shows the dynamic part of the video. The detailed procedure is mentioned in the text I shared on the slack.
204742 [SSV2 dataset]
Comments 🚀
78469 [SSV2 dataset]
Comments 🚀
62041 [SSV2 dataset]
Comments 🚀
212262 [SSV2 dataset]
Comments 🚀
April 09 brush hair u nm np1 ba goo 1 [HMDB51 dataset]
Comments 🚀
HAND STAND CONTEST ! handstand f cm np1 fr med 1 [HMDB51 dataset]
Comments if any 🚀
Big League Chew chew h nm np1 fr goo 0 [HMDB51 dataset]
Comments if any 🚀
50 FIRST DATES punch f nm np1 ri med 15 [HMDB51 dataset]