Part 2, 2021 course?

Part 1 (2020)
65

Per Jeremy’s posts

6 Likes
66

Thanks Moody, that’s very helpful.

It’s hard to follow fast.ai across all their communication channels!

4 Likes
67

Thank you so much for all the updates Moody!!
Truly appreciate it.

1 Like
68

Looking forward to this course and the subsequent part 2.

1 Like
69

I totally agree! Why aren’t anouncements made on the forum? I do not follow twitter. And there is the discord channel as well. I do not see any sense in using all those different medias…

2 Likes
70

Save the time if you want to attend the course live.

2 Likes
71

Is there any information released yet for the remote version like in past years?

Don’t want to miss out on Jeremy explaining transformers.

73

Is there no free online offering this year?

2 Likes
74

I had signed up for the course news and it seems they’re accepting applications (in-person AND remote). The fee is 575 AUD for standard learners and something called “concession” is 175 AUD … the course will be from 6:00pm to 8:00 pm Australian Eastern Time.

In his tweet Jeremy had mentioned that some info might be posted to the fast.ai forums but I have not seen anything “official” as of yet.

Since this course is being offered by a third party and not fast.ai per se, I’m not sure if the videos would (at all) be made public like they used to do for the previous years when it was delivered in San Francisco etc.

3 Likes
75

Jeremy has tweeted that the videos will be made public.

6 Likes
76

Why TF, you can look at icevision, its a modern CV library build on fastai. Unfortunately fastai lacks modern CV tools to tackle instance segmentation or use modern models.
Or you could learn Pytorch Lightning. It’s not like fastai is the only PyTorch library.

77

There’s an extension that does this, SemTorch SemTorch · PyPI

2 Likes
78

Are there videos available? On the website of the course, I can see only slides

79

The “Stat 157” course does seem to have the videos hosted on youtube and linked. They are embeds and at first glance didn’t show for me I had to resize my chrome window a bit.

https://c.d2l.ai/berkeley-stat-157/

The full playlist by Alex Smola is here :

1 Like
80

Thanks, the lib looks interesting, but on the GitHub it says:

just Semantic Segmentation accepted for now

It also seems the last commit was 1 year ago, so I doubt it will be updated.
Apart from the icevision library, I dont know any other library doing instance segmentation with fastai backbone.

81

I am and have been using Semtorch for a year now, and it works fairly well. Do install it from Github though, not from Pypi. Semantic segmentation only, no instance segmentation.

1 Like
82

I would love to learn Pytorch. I’m all for learning more frameworks. I went with TF because of the high quality of video tutorials, reference documentation, and blogs you can easily find. Hence, easier to not get stuck on a problem. Plus after a short bit, it clicked with my brain easier than fast.ai ever did.
Do you have any recommendations for good video tutorials for learning it? paid are okay, if the quality is high.

I’ll look in to Icevision, thanks for the tip.

83

I am also waiting for the 2nd part.

84

I do not remember where I saw it. But there is an approach that makes one-object instance segmentation out of semantic segmentation by using three (or more, I can’t remember right now) classes that are related to pixels on the object, on the background or inside the border area. Then you need a post-processing step (conventional connected components or similar) to convert all the on-object pixels to a number of the instance they belong to.

85

Can someone ask Jeremy about when the part 2 course starts? I will be attending a conference, so I won’t attend it live. :pray:

1 Like