If you have any copyright issues on video, please send us an email at khawar512@gmail.com
0:00 Introduction
1:21 What is Flamingo?
2:20 Flamingo comes in many colours depending on what it cats
4:04 Why did we build it?
5:38 Vision and Language related work
8:16 How does it work?
9:38 Visual processing
11:29 Leveraging an existing language model
13:20 Deal with interleaved visual and text sequence
14:58 The Flamingo family
16:58 Main results: Few-shot results
17:59 On the importance of the training data
18:44 On the importance of freezing the language model
21:45 Adapting to visual dialogue with a simple prompt
23:58 Limitations
26:18 Summary and conclusion
Негізгі бет Transformer for VS | Flamingo: a Visual Language Model for Few-Shot Learning | Session 5 | CVPR 2022
No video
Пікірлер