Non square image rasterization for meshes

Summary:
There are a couple of options for supporting non square images:
1) NDC stays at [-1, 1] in both directions with the distance calculations all modified by (W/H). There are a lot of distance based calculations (e.g. triangle areas for barycentric coordinates etc) so this requires changes in many places.
2) NDC is scaled by (W/H) so the smallest side has [-1, 1]. In this case none of the distance calculations need to be updated and only the pixel to NDC calculation needs to be modified.

I decided to go with option 2 after trying option 1!

API Changes:
- Image size can now be specified optionally as a tuple

TODO:
- add a benchmark test for the non square case.

Reviewed By: jcjohnson

Differential Revision: D24404975

fbshipit-source-id: 545efb67c822d748ec35999b35762bce58db2cf4

docs/notes/renderer_getting_started.md

@@ -55,6 +55,19 @@ While we tried to emulate several aspects of OpenGL, there are differences in th
 
 ---
 
+### Rasterizing Non Square Images
+
+To rasterize an image where H != W, you can specify the `image_size` in the `RasterizationSettings` as a tuple of (H, W).
+
+The aspect ratio needs special consideration. There are two aspect ratios to be aware of:
+    - the aspect ratio of each pixel
+    - the aspect ratio of the output image
+In the cameras e.g. `FoVPerspectiveCameras`, the `aspect_ratio` argument can be used to set the pixel aspect ratio. In the rasterizer, we assume square pixels, but variable image aspect ratio (i.e rectangle images).
+
+In most cases you will want to set the camera aspect ratio to 1.0 (i.e. square pixels) and only vary the `image_size` in the `RasterizationSettings`(i.e. the output image dimensions in pixels).
+
+---
+
 ### The pulsar backend
 
 Since v0.3, [pulsar](https://arxiv.org/abs/2004.07484) can be used as a backend for point-rendering. It has a focus on efficiency, which comes with pros and cons: it is highly optimized and all rendering stages are integrated in the CUDA kernels. This leads to significantly higher speed and better scaling behavior. We use it at Facebook Reality Labs to render and optimize scenes with millions of spheres in resolutions up to 4K. You can find a runtime comparison plot below (settings: `bin_size=None`, `points_per_pixel=5`, `image_size=1024`, `radius=1e-2`, `composite_params.radius=1e-4`; benchmarked on an RTX 2070 GPU).
@@ -75,6 +88,8 @@ For mesh texturing we offer several options (in `pytorch3d/renderer/mesh/texturi
 
 <img src="assets/texturing.jpg" width="1000">
 
+---
+
 ### A simple renderer
 
 A renderer in PyTorch3D is composed of a **rasterizer** and a **shader**. Create a renderer in a few simple steps:
@@ -108,6 +123,8 @@ renderer = MeshRenderer(
 )
 ```
 
+---
+
 ### A custom shader
 
 Shaders are the most flexible part of the PyTorch3D rendering API. We have created some examples of shaders in `shaders.py` but this is a non exhaustive set.