I have transitioned the Quickdraw starter pack to now use the dataset API. There were models I wanted to train where I would need to hack together custom Datasets and potentially custom Dataloaders - the dataset API makes the headaches go away
As for the starter pack, this time I ironed out a couple of the rough edges of the earlier version. I also now generate the drawings on the fly so experimentation should be much easier now.
The only annoying thing about this dataset is how long training takes with size 256x256 - but maybe there is a way to get equally good results with smaller sizes?!
Please check out my work on Optimizing Image-Classification using Transfer-Learning! This is an image classifier of 4 different types of Arctic Dogs! This medium blog tells you step by step of how I finally bring down the error rate at the end after a few tips and tricks from our previous lesson 3 lecture.
I like the idea that you are incorporating users feedback into your next training iteration. However, you do want to put manual inspection in between because feedback is not always right
Talking about collaborative filtering, I’ve created a small post when was watching the previous version of the course. The code from the post is written in PyTorch but probably could be interesting for someone who wants to dig deeper into the topic.
One of the gists from the post that shows writing of a small custom nn.Module with embeddings:
Update 1: Not sure why the link to Medium is not rendered properly, here is a plain address:
Otherwise, you probably could find it via @iliazaitsev username on Medium.
Update 2: Ok, Medium support responded that my account was blocked automatically by their spam filter. Probably they need try some Deep Learning methods to reduce the number of false positives
Hi Jon. The latest update to the API is quite different from the code in my post so I’ll hopefully refactor it this weekend and link to the new notebook at the bottom.
The dataset was created using some software, or maybe a camera/device that gives out these key points. Whatever maybe, there must be a underlying mathematical model(fn) for that (camera/device/software). So, that’s what the neural network is trying to approximate, instead of the finding the actual key points. What I mean to say is, here the neural network is not trying to find exactly where is the mouth, eyes or nose; because we haven’t explicitly mentioned it in our dataset. Yoshua Bengio and team created this dataset, I would like to know if there was any intention of such sorts.
if that’s the case, even if we predict the actual facial keypoints for the test set, we can expect a higher error.
Notebook is still a work in progress. I’ll share a clean version soon.
Also, The submission file is a bit weird. I’m not sure why they are not evaluating on the basis of all the points.
I’ve been using ZEIT but it’s slow as heck when waking up the app from a frozen state. Impressed by how responsive your emoji app is and so interested on what you all are doing.
I’ve made lynx classifier (it classifies which lynx species is given lynx). https://which-lynx-is-it.now.sh/
Error is something 20%ish (lost the notebook, because of issue with gcp). Considering the fact that the dataset was noisy it think it’s good. Interesting thing is it really has problem classifying baby lynxes for some reason.
Hopefully I will have time this week to write a blog post about it.
It’s just a small ec2 instance (behind an ELB to terminate the SSL which is required for WebRTC to work).
I was really surprised at how much throughput we could get during fastai inference. The fact that we were sending smaller cropped images from the client really helped.
I’ve written the following short Medium post doing some theory review of the concepts we deepened into during Lesson 3, like the learning rate and activation functions.
