[Paper Review] Unsupervised Learning of Visual Representations using Videos(2015)

Unsupervised Learning of Visual Representations using Videos

Xiaolong Wang, Abhinav Gupta, arXiv 2015

 

PDF, Video By SeonghoonYu July 23th, 2021

 

Summary

 This paper use hundreds of thousands of unlabeled videos from the web to learn visual representations.

 

 

 They use the first frame and the last frame in same video as positive samples and a random frame from different video as negative sample. They train Siamese Triplet Network sharing same paremeters on these positive sample and negative sample by minimizing triplet loss. In this way, positive samples are closer each other as compared to negatives samples.

 

 

cosine distance between two samples

 

triplet loss, M is margin

 

Loss function

 

Experiment

 Achieve the performance closed to its supervised counterpats

 

 

What I like about the paper

• They use a number of unlabeled video datas to train the model on unsupervised learning
• achieves the performance closed to its supervised counterparts

---

my github about what i read

Seonghoon-Yu/Paper_Review_and_Implementation_in_PyTorch