14 پاسخ به “MIT 6.S094: چشم انداز رایانه”

1. Thank you Lex, very clear and informative.
   One thing: 3.59 "the structure of the visual cortex is in layers". The slide before, differences, topology: human brains have no "layers" — a bit misleading

2. be yourself man

3. So much to unpack here, thank you for uploading !

4. # Some timestamps:
   – Computer Vision is Deep Learning: 00:53 – 01:57
   – Images are numbers: 01:58 – 03:13
   – Visual Cortex: 03:51 – 04:37
   – Deep learning is hard: 04:38 – 06:42
   – KNN: 10:44 – 11:32
   – Learning is optimization of a function: 12:05 – 12:42
   – CNN: 13:36 – 21:30
   – SENet: 29:30 – 32:06
   – Capsule Networks: 32:07 – 34:23
   – Fully Convolutional Networks: 35:22 – 37:53
   – SegNet: 39:50 – 40:15
   – Dilated convolutions: 40:16 – 41:05
   – Deeplap: 41:06 – 41:49
   – Key Aspects of Segment: 41:50 – 42:50
   – ResNet-DUC
   – FlowNet: 44:26 – 48:47
   – FlowNet 2.0: 48:49 – 50:06

5. The way Lex speaks is very powerful. Always to the point and every sentence have a meaning and purpose with a clear voice. Love it. Love the video. Liked and subscribed

6. These lectures are amazing.

7. Dear Lex, thanks so much for providing such good series of videos. For the capsule networks and GAN mentions at last of this lecture? Where could we find the them?

8. The more I look at different kinds of networks and their strengths and weaknesses, the more I believe that unsupervised learning for relabeling used during yak scrapping is an important tool. In the case given for capsule networks, i.e. the CNN discounting the relationship information, a simple relabeling would guarantee that the CNN would properly identify disarranged faces. Unsupervised clustering would be a good way to identify what relabeling needs to occur.

9. Thank you very very much!

10. Hey Lex. Thank you for great lectures. are GAN and CapsNet available online?

11. Hey Lex, are you going to be uploading the lectures on capsule networks?

12. Awesome video

13. Very happy to see you will be dedicating time to the capsule networks. They are intriguing! Looking forward to it.

14. In a typical CNN, we slide the 'window' from left to right, top to bottom. Is this a Western bias, derived from the way we used to read scrolled paper ?

    The left-right-top-bottom rectangular scan doesn't have any intrinsic meaning, and doesn't mimic the way our retinas work.

    What happens if we use circular rather than rectangular images, and 'grow' from the centre of an image outward ?

نظرات بسته شده اند.