Lr_find, fit_one_cycle questions & Learning resources

Part 1 (2019)
1

I had a few questions:

  1. Does lr_find and fit_one_cycle work with optimizers other than SGD with momentum such as Adam, AdamW, RMSProp. Can someone please give a brief explanation.

  2. How does one get started with object detection? I noticed that this topic was not covered in the course. I am kinda fascinated with this topic. Could someone be kind as to point me towards the relevant resources. I know about the different APIs like detectron, tensorflow object detection but i wanted to start from scratch at first :blush:

Thanks

2

hey Ayushman,
1- lr_find and fit_one_cycle both train the model with Adam as default (you can also find the answer by looking at source code or documentation )
2- I’m also interested in that field but as far as I know course v3 is not focused on that but you can take a look at lesson 8 and 9 of the 2018 version (v2) of the course. You can find the notebooks in the forum if you search. Also videos are available on Jeremy’s YouTube channel
Regards

1 Like
3

I believe part 2 gets onto object detection. I don’t see it now but I know I’ve seen a video somewhere. Hunt around.

4

Replying in tandem, that’s probably where I saw them, now I think about it I did just see the vid in my YouTube feed.

5

part 2 of v3? I haven’t watched the last two videos yet (which are taught with Swift). maybe in those two lesson :thinking:

6

:ok_hand: :ok_hand:

7

I found the videos thanks…

1 Like
8

I’d also potentially look at @lgvaz’s Mantis Shrimp library

https://lgvaz.github.io/mantisshrimp/tutorials/getting_started.html

It tries to merge fastai2 w/ object detection (though had to rebuild a few things) and the docs are very good :slight_smile:

2 Likes