fix default settings for point rasterization and update benchmark

Summary:
Fixes the default setting of `max_points_per_bin` in `rasterize_points.py`. For large batches with large size pointclouds this was a causing the rasterizer to be very slow.

Expanded the pointcloud rendering benchmarks to include larger size pointclouds and fixed cuda synchronization issue in benchmark.

Reviewed By: gkioxari

Differential Revision: D22301185

fbshipit-source-id: 5077c1ba2c43d73efc1c659f0ec75959ceddf893

pytorch3d/renderer/points/rasterize_points.py

@@ -98,7 +98,7 @@ def rasterize_points(
             )
 
     if max_points_per_bin is None:
-        max_points_per_bin = int(max(10000, points_packed.shape[0] / 5))
+        max_points_per_bin = int(max(10000, pointclouds._P / 5))
 
     # Function.apply cannot take keyword args, so we handle defaults in this
     # wrapper and call apply with positional args only

tests/bm_rasterize_points.py

@@ -28,10 +28,17 @@ def _bm_cpu_with_init(N, P, img_size=32, radius=0.1, pts_per_pxl=3):
 
 def _bm_cuda_with_init(N, P, img_size=32, radius=0.1, pts_per_pxl=3):
     torch.manual_seed(231)
-    points = torch.randn(N, P, 3, device=torch.device("cuda"))
+    device = torch.device("cuda:0")
+    points = torch.randn(N, P, 3, device=device)
     pointclouds = Pointclouds(points=points)
     args = (pointclouds, img_size, radius, pts_per_pxl)
-    return lambda: rasterize_points(*args)
+    torch.cuda.synchronize(device)
+
+    def fn():
+        rasterize_points(*args)
+        torch.cuda.synchronize(device)
+
+    return fn
 
 
 def bm_python_vs_cpu() -> None:
@@ -46,4 +53,9 @@ def bm_python_vs_cpu() -> None:
         {"N": 4, "P": 1024, "img_size": 128, "radius": 0.05, "pts_per_pxl": 5},
     ]
     benchmark(_bm_cpu_with_init, "RASTERIZE_CPU", kwargs_list, warmup_iters=1)
+    kwargs_list += [
+        {"N": 32, "P": 10000, "img_size": 128, "radius": 0.01, "pts_per_pxl": 50},
+        {"N": 32, "P": 100000, "img_size": 128, "radius": 0.01, "pts_per_pxl": 50},
+        {"N": 8, "P": 200000, "img_size": 512, "radius": 0.01, "pts_per_pxl": 50},
+    ]
     benchmark(_bm_cuda_with_init, "RASTERIZE_CUDA", kwargs_list, warmup_iters=1)