From ac915dfad8754ae3717f9428a412ac069755b116 Mon Sep 17 00:00:00 2001 From: Shauray Singh <39147312+shauray8@users.noreply.github.com> Date: Fri, 18 Jun 2021 23:34:28 +0530 Subject: [PATCH] Update README.md --- README.md | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) diff --git a/README.md b/README.md index 59df527..6def141 100644 --- a/README.md +++ b/README.md @@ -12,8 +12,9 @@ the goal is to predict the direction of travel (in camera frame) from provided d The example labels are generated using a Neural Network, and the labels were confirmed with a SLAM algorithm. You can estimate the focal length to be 910 pixels. -![](./Docs/yaw-pitch-roll.png) - +
+ +
## Evaluation They will evaluate our mean squared error against our ground truth labels. Errors for frames where the car speed is less than 4m/s will be ignored. Those are also labelled as NaN in the example labels. @@ -38,9 +39,9 @@ but I don't think it makes a lot of difference. ![](./Docs/FlowNetARch.png) after a bunch of ConvNets, it goes through a refinement layer the output for the above architecture is the input for the refinement layer! - -![](./Docs/FlowNetRef.png) - ++ +
This pretty much summarizes the architecture and at the end rather than implementing the last layer I make the matrix pass through a Linear layer and predict yaw and pitch with ONE HOT vector kinda thing. If you have a better idea for the ONE HOT vector alternative just let me know !!