Part 2 Lesson 8 wiki

(ecdrid) #53

Wrote a answer on Stack - DataScience explaining Transfer Learning ( thanks to Jeremy)

Read it here if you like…

0 Likes

(Matt Trent) #58

Can you share the resource you saw the PCI-e profiling of x8 vs x16?

2 Likes

(Vikas Bahirwani) #59

Here’s a good link for Greek Alphabets . Please reply if you have a better link.

2 Likes

(Kartik Godawat) #60

This one contains a couple of benchmarks run on TitanX

2 Likes

(Vikas Bahirwani) #61

Jeremy mentioned developing parts of this course on Windows.

I would love references to his Windows Setup (pytorch and fastai libraries on Windows?).

Thanks
Vikas

0 Likes

(Emil) #64

Here is a quick summary for downloading the dataset:

cd ~/fastai/courses/dl2
ln -s ~/data data && cd $_
mkdir pascal && cd $_
curl -OL http://pjreddie.com/media/files/VOCtrainval_06-Nov-2007.tar
curl -OL https://storage.googleapis.com/coco-dataset/external/PASCAL_VOC.zip
tar -xf VOCtrainval_06-Nov-2007.tar
unzip PASCAL_VOC.zip
mv PASCAL_VOC/*.json .
rmdir PASCAL_VOC
15 Likes

Deep Learning Brasília - Lição 8
(Jaya) #65

Can one access the pascal dataset from the website instead of downloading it? Can Path() do it?

0 Likes

(Matt Trent) #67

Why did you choose upper-left + lower-right instead of upper-left + dimensions?

1 Like

(unknown) #68

Thank you for posting this. Where did you get the google link with the .Json files? I seemed to have missed that one.

0 Likes

(Emil) #69

Great blog post about object-oriented matplotlib API: https://realpython.com/blog/python/python-matplotlib-guide/

18 Likes

(unknown) #70

Found it, it is after the PATH command. Still wondering where the original xml files are though

0 Likes

(KRO) #71

In case anyone likes to solidify things with a song, you can sing the greek alphabet to the tune of the song “row row row your boat”. :slight_smile:

2 Likes

(Emil) #72

Actually, there is a link in one of the Markdown cells with description, after list(PATH.iterdir()). It’s easily missable. I, too, didn’t see it immediately.

1 Like

(Ankit Goila) #73

I believe working with/specifying 2 parameters is less cumbersome than 3?

0 Likes

(Pavel Surmenok) #74

Does using bottom-left corner instead of width/height of the box impact accuracy of the model?

0 Likes

(chunduri) #75

How is bounding box identified in this example. Which algorithm is used, ie YOLO, R-CNN?

0 Likes

(unknown) #76

It shouldn’t assuming that you stick to the same precision (if bound by pixels, stay to that). It is just a different way to defining the same thing.

1 Like

(Aza Raskin) #77

Given the move towards differentiable programming, why not use complex-valued weights instead of real-valued weights?

Complex functions, if differentiable, are infinitely differentiable—and complex numbers represent rotation and scaling more succinctly than real numbers and so should require fewer parameters in a model?

5 Likes

(Maurice Kroon) #79

Does anyone know the name of the extension for atom that does the search function Jeremy just did in VSC?
So far I’ve been using ‘find in folder’ but this functionality looks pretty neat.

2 Likes

(Emil) #82

Now it’s just predicting the same class that the largest bounding box has for a particular image, with resnet34.

Later in the notebook, they just run a regression with 4 outputs for x1, y1, x2 and y2, again with the same resnet34.

Even later, they combine classification and regression in a single model with 4 + number_of_categories outputs and create a custom loss that breaks the outputs into classification and regression parts and applies L1 and cross-entropy losses to these parts.

10 Likes