From 2f379bd77f6d2445b6d80f1d6b0ae1115e34a81c Mon Sep 17 00:00:00 2001 From: Varun Jewalikar Date: Mon, 11 Jul 2016 12:51:28 +0100 Subject: [PATCH] Fix typo Changed 'scale' -> 'scaled' in the phrase "with no ideas about how this could be possibly scale". Could also be corrected as "with no ideas about how this could possibly scale". --- _posts/2014-07-03-feature-learning-escapades.markdown | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/_posts/2014-07-03-feature-learning-escapades.markdown b/_posts/2014-07-03-feature-learning-escapades.markdown index 871e63613..2689b14f6 100644 --- a/_posts/2014-07-03-feature-learning-escapades.markdown +++ b/_posts/2014-07-03-feature-learning-escapades.markdown @@ -54,7 +54,7 @@ This was also around the time when the Kinect came out, so I thought I'd give 3D 1. There is no obvious/clean way to plug a neural network into 3D data. 2. Reasoning about the difference between occluded / empty space is a huge pain. -3. It is very hard to collect data at scale. Neural nets love data and here I was playing around with datasets on order of 100 scenes, with no ideas about how this could be possibly scale. +3. It is very hard to collect data at scale. Neural nets love data and here I was playing around with datasets on order of 100 scenes, with no ideas about how this could be possibly scaled. 4. I was working with fully static 3D environments. No movement, no people, no fun. I ended up doing a bit of Unsupervised Object Discovery in my 3D meshes and publishing it at a robotics conference, where it was most relevant (Object Discovery in 3D scenes via Shape Analysis). I was happy that I found a very simple, efficient and surprisingly effective way of computing objectness over 3D meshes, but it wasn't what I set out to do. I followed up on the project a bit while working with Sebastian Thrun for my last rotation, but I remained unsatisfied and unfulfilled. There was no brain stuff, no huge datasets to learn from, and even if it all worked, it would work on static, boring scenes.